AI/rocm-systems
2
0
Forka 0
Codice Problemi Pull Request Actions Pacchetti Progetti Rilasci Wiki Attività

merge from master

Sfoglia il codice sorgente
This commit is contained in:
Anusha Godavarthy Surya
2019-10-25 15:52:09 +05:30
parent 259d8b4cdf 70f2cd1317
commit 5f47e99ffe
40 ha cambiato i file con 1341 aggiunte e 690 eliminazioni
Scarica il file Patch Scarica il file Diff Expand all files Collapse all files
include/hip/hcc_detail/code_object_bundle.hpp
+1 -1
Vedi File
@@ -86,7 +86,7 @@ struct Bundled_code {
char cbuf[sizeof(offset) + sizeof(bundle_sz) + sizeof(triple_sz)];
} header;
std::string triple;
std::vector<char> blob;
std::string blob;
};
#define magic_string_ "__CLANG_OFFLOAD_BUNDLE__"
include/hip/hcc_detail/functional_grid_launch.hpp
-29
Vedi File
@@ -127,35 +127,6 @@ void hipLaunchKernelGGLImpl(
} // Namespace hip_impl.
template <typename F>
inline
hipError_t hipOccupancyMaxPotentialBlockSize(uint32_t* gridSize, uint32_t* blockSize,
F kernel, size_t dynSharedMemPerBlk, uint32_t blockSizeLimit) {
using namespace hip_impl;
hip_impl::hip_init();
auto f = get_program_state().kernel_descriptor(reinterpret_cast<std::uintptr_t>(kernel),
target_agent(0));
return hipOccupancyMaxPotentialBlockSize(gridSize, blockSize, f,
dynSharedMemPerBlk, blockSizeLimit);
}
template <typename F>
inline
hipError_t hipOccupancyMaxActiveBlocksPerMultiprocessor(uint32_t* numBlocks, F kernel,
uint32_t blockSize, size_t dynSharedMemPerBlk) {
using namespace hip_impl;
hip_impl::hip_init();
auto f = get_program_state().kernel_descriptor(reinterpret_cast<std::uintptr_t>(kernel),
target_agent(0));
return hipOccupancyMaxActiveBlocksPerMultiprocessor(numBlocks, f, blockSize, dynSharedMemPerBlk);
}
template <typename... Args, typename F = void (*)(Args...)>
inline
void hipLaunchKernelGGL(F kernel, const dim3& numBlocks, const dim3& dimBlocks,
include/hip/hcc_detail/hip_atomic.h
+32 -10
Vedi File
@@ -59,12 +59,17 @@ float atomicAdd(float* address, float val)
{
unsigned int* uaddr{reinterpret_cast<unsigned int*>(address)};
unsigned int old{__atomic_load_n(uaddr, __ATOMIC_RELAXED)};
unsigned int r;
unsigned int r;
do {
r = old;
r = __atomic_load_n(uaddr, __ATOMIC_RELAXED);
if (r != old) { r = old; continue; }
old = atomicCAS(uaddr, r, __float_as_uint(val + __uint_as_float(r)));
} while (r != old);
if (r == old) break;
} while (true);
return __uint_as_float(r);
}
@@ -74,13 +79,18 @@ double atomicAdd(double* address, double val)
{
unsigned long long* uaddr{reinterpret_cast<unsigned long long*>(address)};
unsigned long long old{__atomic_load_n(uaddr, __ATOMIC_RELAXED)};
unsigned long long r;
unsigned long long r;
do {
r = old;
r = __atomic_load_n(uaddr, __ATOMIC_RELAXED);
if (r != old) { r = old; continue; }
old = atomicCAS(
uaddr, r, __double_as_longlong(val + __longlong_as_double(r)));
} while (r != old);
if (r == old) break;
} while (true);
return __longlong_as_double(r);
}
@@ -144,7 +154,13 @@ unsigned long long atomicMin(
unsigned long long* address, unsigned long long val)
{
unsigned long long tmp{__atomic_load_n(address, __ATOMIC_RELAXED)};
while (val < tmp) { tmp = atomicCAS(address, tmp, val); }
while (val < tmp) {
const auto tmp1 = __atomic_load_n(address, __ATOMIC_RELAXED);
if (tmp1 != tmp) { tmp = tmp1; continue; }
tmp = atomicCAS(address, tmp, val);
}
return tmp;
}
@@ -167,7 +183,13 @@ unsigned long long atomicMax(
unsigned long long* address, unsigned long long val)
{
unsigned long long tmp{__atomic_load_n(address, __ATOMIC_RELAXED)};
while (tmp < val) { tmp = atomicCAS(address, tmp, val); }
while (tmp < val) {
const auto tmp1 = __atomic_load_n(address, __ATOMIC_RELAXED);
if (tmp1 != tmp) { tmp = tmp1; continue; }
tmp = atomicCAS(address, tmp, val);
}
return tmp;
}
@@ -177,7 +199,7 @@ inline
unsigned int atomicInc(unsigned int* address, unsigned int val)
{
__device__
extern
extern
unsigned int __builtin_amdgcn_atomic_inc(
unsigned int*,
unsigned int,
@@ -194,7 +216,7 @@ inline
unsigned int atomicDec(unsigned int* address, unsigned int val)
{
__device__
extern
extern
unsigned int __builtin_amdgcn_atomic_dec(
unsigned int*,
unsigned int,
include/hip/hcc_detail/hip_runtime_api.h
+36 -16
Vedi File
@@ -1742,10 +1742,10 @@ hipError_t hipMemset(void* dst, int value, size_t sizeBytes);
*
* @param[out] dst Data ptr to be filled
* @param[in] constant value to be set
* @param[in] sizeBytes Data size in bytes
* @param[in] number of values to be set
* @return #hipSuccess, #hipErrorInvalidValue, #hipErrorNotInitialized
*/
hipError_t hipMemsetD8(hipDeviceptr_t dest, unsigned char value, size_t sizeBytes);
hipError_t hipMemsetD8(hipDeviceptr_t dest, unsigned char value, size_t count);
/**
* @brief Fills the first sizeBytes bytes of the memory area pointed to by dest with the constant
@@ -1758,11 +1758,11 @@ hipError_t hipMemsetD8(hipDeviceptr_t dest, unsigned char value, size_t sizeByte
*
* @param[out] dst Data ptr to be filled
* @param[in] constant value to be set
* @param[in] sizeBytes Data size in bytes
* @param[in] number of values to be set
* @param[in] stream - Stream identifier
* @return #hipSuccess, #hipErrorInvalidValue, #hipErrorNotInitialized
*/
hipError_t hipMemsetD8Async(hipDeviceptr_t dest, unsigned char value, size_t sizeBytes, hipStream_t stream __dparm(0));
hipError_t hipMemsetD8Async(hipDeviceptr_t dest, unsigned char value, size_t count, hipStream_t stream __dparm(0));
/**
* @brief Fills the first sizeBytes bytes of the memory area pointed to by dest with the constant
@@ -1770,10 +1770,10 @@ hipError_t hipMemsetD8Async(hipDeviceptr_t dest, unsigned char value, size_t siz
*
* @param[out] dst Data ptr to be filled
* @param[in] constant value to be set
* @param[in] sizeBytes Data size in bytes
* @param[in] number of values to be set
* @return #hipSuccess, #hipErrorInvalidValue, #hipErrorNotInitialized
*/
hipError_t hipMemsetD16(hipDeviceptr_t dest, unsigned short value, size_t sizeBytes);
hipError_t hipMemsetD16(hipDeviceptr_t dest, unsigned short value, size_t count);
/**
* @brief Fills the first sizeBytes bytes of the memory area pointed to by dest with the constant
@@ -1786,11 +1786,11 @@ hipError_t hipMemsetD16(hipDeviceptr_t dest, unsigned short value, size_t sizeBy
*
* @param[out] dst Data ptr to be filled
* @param[in] constant value to be set
* @param[in] sizeBytes Data size in bytes
* @param[in] number of values to be set
* @param[in] stream - Stream identifier
* @return #hipSuccess, #hipErrorInvalidValue, #hipErrorNotInitialized
*/
hipError_t hipMemsetD16Async(hipDeviceptr_t dest, unsigned short value, size_t sizeBytes, hipStream_t stream __dparm(0));
hipError_t hipMemsetD16Async(hipDeviceptr_t dest, unsigned short value, size_t count, hipStream_t stream __dparm(0));
/**
* @brief Fills the memory area pointed to by dest with the constant integer
@@ -2884,14 +2884,14 @@ hipError_t hipLaunchCooperativeKernelMultiDevice(hipLaunchParams* launchParamsLi
* @param [out] gridSize minimum grid size for maximum potential occupancy
* @param [out] blockSize block size for maximum potential occupancy
* @param [in] f kernel function for which occupancy is calulated
* @param [in] dynSharedMemPerBlk dynamic shared memory usage (in bytes) intended for each block
* @param [in] dynamicSMemSize Per - block dynamic shared memory usage intended, in bytes
* @param [in] blockSizeLimit the maximum block size for the kernel, use 0 for no limit
*
* @returns hipSuccess, hipInvalidDevice, hipErrorInvalidValue
*/
hipError_t hipOccupancyMaxPotentialBlockSize(uint32_t* gridSize, uint32_t* blockSize,
hipFunction_t f, size_t dynSharedMemPerBlk,
uint32_t blockSizeLimit);
hipError_t hipOccupancyMaxPotentialBlockSize(int* gridSize, int* blockSize,
const void* f, size_t dynamicSMemSize,
int blockSizeLimit);
/**
* @brief Returns occupancy for a device function.
@@ -2899,10 +2899,10 @@ hipError_t hipOccupancyMaxPotentialBlockSize(uint32_t* gridSize, uint32_t* block
* @param [out] numBlocks Returned occupancy
* @param [in] func Kernel function for which occupancy is calulated
* @param [in] blockSize Block size the kernel is intended to be launched with
* @param [in] dynSharedMemPerBlk dynamic shared memory usage (in bytes) intended for each block
* @param [in] dynamicSMemSize Per - block dynamic shared memory usage intended, in bytes
*/
hipError_t hipOccupancyMaxActiveBlocksPerMultiprocessor(
uint32_t* numBlocks, hipFunction_t f, uint32_t blockSize, size_t dynSharedMemPerBlk);
int* numBlocks, const void* f, int blockSize, size_t dynamicSMemSize);
/**
* @brief Returns occupancy for a device function.
@@ -2910,11 +2910,11 @@ hipError_t hipOccupancyMaxActiveBlocksPerMultiprocessor(
* @param [out] numBlocks Returned occupancy
* @param [in] func Kernel function for which occupancy is calulated
* @param [in] blockSize Block size the kernel is intended to be launched with
* @param [in] dynSharedMemPerBlk dynamic shared memory usage (in bytes) intended for each block
* @param [in] dynamicSMemSize Per - block dynamic shared memory usage intended, in bytes
* @param [in] flags Extra flags for occupancy calculation (currently ignored)
*/
hipError_t hipOccupancyMaxActiveBlocksPerMultiprocessorWithFlags(
uint32_t* numBlocks, hipFunction_t f, uint32_t blockSize, size_t dynSharedMemPerBlk, unsigned int flags);
int* numBlocks, const void* f, int blockSize, size_t dynamicSMemSize, unsigned int flags);
/**
* @brief Launches kernels on multiple devices and guarantees all specified kernels are dispatched
@@ -3320,7 +3320,27 @@ hipError_t hipBindTextureToMipmappedArray(const texture<T, dim, readMode>& tex,
return hipSuccess;
}
template <class T>
inline hipError_t hipOccupancyMaxActiveBlocksPerMultiprocessor(
int* numBlocks, T f, int blockSize, size_t dynamicSMemSize) {
return hipOccupancyMaxActiveBlocksPerMultiprocessor(
numBlocks, reinterpret_cast<const void*>(f), blockSize, dynamicSMemSize);
}
template <class T>
inline hipError_t hipOccupancyMaxActiveBlocksPerMultiprocessorWithFlags(
int* numBlocks, T f, int blockSize, size_t dynamicSMemSize, unsigned int flags) {
return hipOccupancyMaxActiveBlocksPerMultiprocessorWithFlags(
numBlocks, reinterpret_cast<const void*>(f), blockSize, dynamicSMemSize, flags);
}
template <class T>
inline hipError_t hipOccupancyMaxPotentialBlockSize(int* gridSize, int* blockSize,
T f, size_t dynamicSMemSize, int blockSizeLimit) {
return hipOccupancyMaxPotentialBlockSize(
gridSize, blockSize, reinterpret_cast<const void*>(f), dynamicSMemSize, blockSizeLimit);
}
template <class T>
inline hipError_t hipLaunchCooperativeKernel(T f, dim3 gridDim, dim3 blockDim,
void** kernelParams, unsigned int sharedMemBytes, hipStream_t stream) {
include/hip/hcc_detail/hip_runtime_prof.h
+77
Vedi File
@@ -0,0 +1,77 @@
/*
Copyright (c) 2019 - present 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_HCC_DETAIL_HIP_RUNTIME_PROF_H
#define HIP_INCLUDE_HIP_HCC_DETAIL_HIP_RUNTIME_PROF_H
// HIP VDI Op IDs enumeration
enum HipVdiOpId {
kHipVdiOpIdDispatch = 0,
kHipVdiOpIdCopy = 1,
kHipVdiOpIdBarrier = 2,
kHipVdiOpIdNumber = 3
};
// Types of VDI commands
enum HipVdiCommandKind {
kHipVdiCommandKernel = 0x11F0,
kHipVdiMemcpyDeviceToHost = 0x11F3,
kHipHipVdiMemcpyHostToDevice = 0x11F4,
kHipVdiMemcpyDeviceToDevice = 0x11F5,
kHipVidMemcpyDeviceToHostRect = 0x1201,
kHipVdiMemcpyHostToDeviceRect = 0x1202,
kHipVdiMemcpyDeviceToDeviceRect = 0x1203,
kHipVdiFillMemory = 0x1207,
};
/**
* @brief Initializes activity callback
*
* @param [input] id_callback Event ID callback function
* @param [input] op_callback Event operation callback function
* @param [input] arg Arguments passed into callback
*
* @returns None
*/
void hipInitActivityCallback(void* id_callback, void* op_callback, void* arg);
/**
* @brief Enables activity callback
*
* @param [input] op Operation, which will trigger a callback (@see HipVdiOpId)
* @param [input] enable Enable state for the callback
*
* @returns True if successful
*/
bool hipEnableActivityCallback(uint32_t op, bool enable);
/**
* @brief Returns the description string for the operation kind
*
* @param [input] id Command kind id (@see HipVdiCommandKind)
*
* @returns A pointer to a const string with the command description
*/
const char* hipGetCmdName(uint32_t id);
#endif // HIP_INCLUDE_HIP_HCC_DETAIL_HIP_RUNTIME_PROF_H
include/hip/hcc_detail/hip_vector_types.h
+96 -13
Vedi File
@@ -47,6 +47,95 @@ THE SOFTWARE.
#if defined(__cplusplus)
#include <type_traits>
namespace hip_impl {
template<typename T, typename Vector, unsigned int idx>
struct Scalar_accessor {
// Idea from https://t0rakka.silvrback.com/simd-scalar-accessor
Vector data;
__host__ __device__
operator T() const noexcept { return data[idx]; }
__host__ __device__
Scalar_accessor& operator=(T x) noexcept {
data[idx] = x;
return *this;
}
__host__ __device__
Scalar_accessor& operator+=(T x) noexcept {
data[idx] += x;
return *this;
}
__host__ __device__
Scalar_accessor& operator-=(T x) noexcept {
data[idx] -= x;
return *this;
}
__host__ __device__
Scalar_accessor& operator*=(T x) noexcept {
data[idx] *= x;
return *this;
}
__host__ __device__
Scalar_accessor& operator/=(T x) noexcept {
data[idx] /= x;
return *this;
}
template<
typename U = T,
typename std::enable_if<std::is_integral<U>{}>::type* = nullptr>
__host__ __device__
Scalar_accessor& operator%=(T x) noexcept {
data[idx] %= x;
return *this;
}
template<
typename U = T,
typename std::enable_if<std::is_integral<U>{}>::type* = nullptr>
__host__ __device__
Scalar_accessor& operator>>=(T x) noexcept {
data[idx] >>= x;
return *this;
}
template<
typename U = T,
typename std::enable_if<std::is_integral<U>{}>::type* = nullptr>
__host__ __device__
Scalar_accessor& operator<<=(T x) noexcept {
data[idx] <<= x;
return *this;
}
template<
typename U = T,
typename std::enable_if<std::is_integral<U>{}>::type* = nullptr>
__host__ __device__
Scalar_accessor& operator&=(T x) noexcept {
data[idx] &= x;
return *this;
}
template<
typename U = T,
typename std::enable_if<std::is_integral<U>{}>::type* = nullptr>
__host__ __device__
Scalar_accessor& operator|=(T x) noexcept {
data[idx] |= x;
return *this;
}
template<
typename U = T,
typename std::enable_if<std::is_integral<U>{}>::type* = nullptr>
__host__ __device__
Scalar_accessor& operator^=(T x) noexcept {
data[idx] ^= x;
return *this;
}
};
} // Namespace hip_impl.
template<typename T, unsigned int n> struct HIP_vector_base;
template<typename T>
@@ -55,9 +144,7 @@ THE SOFTWARE.
union {
Native_vec_ data;
struct {
T x;
};
hip_impl::Scalar_accessor<T, Native_vec_, 0> x;
};
};
@@ -67,10 +154,8 @@ THE SOFTWARE.
union {
Native_vec_ data;
struct {
T x;
T y;
};
hip_impl::Scalar_accessor<T, Native_vec_, 0> x;
hip_impl::Scalar_accessor<T, Native_vec_, 1> y;
};
};
@@ -238,12 +323,10 @@ THE SOFTWARE.
union {
Native_vec_ data;
struct {
T x;
T y;
T z;
T w;
};
hip_impl::Scalar_accessor<T, Native_vec_, 0> x;
hip_impl::Scalar_accessor<T, Native_vec_, 1> y;
hip_impl::Scalar_accessor<T, Native_vec_, 2> z;
hip_impl::Scalar_accessor<T, Native_vec_, 3> w;
};
};