AI/rocm-systems
2
0
Fork 0
Code Issues Pull Requests Actions Packages Projects Releases Wiki Activity

Merge branch 'master' into amd-master-next

Browse Source 
Conflicts:
	CMakeLists.txt
	tests/src/texture/simpleTexture2DLayered.cpp
	tests/src/texture/simpleTexture3D.cpp

Change-Id: I4aa4754d391b5f37ddf15fa0bcfc84d9da020119
This commit is contained in:
Lad, Aditya
2020-03-06 14:10:04 -05:00
parent 1216153c66 cf61ad2830
commit e7fdb3d796
42 changed files with 1410 additions and 640 deletions
Download Patch File Download Diff File Expand all files Collapse all files
hipamd/include/hip/hcc_detail/code_object_bundle.hpp
+12 -8
View File
@@ -31,9 +31,11 @@ THE SOFTWARE.
#include <string>
#include <utility>
#include <vector>
#include <unordered_set>
namespace hip_impl {
#if !defined(DISABLE_REDUCED_GPU_BLOB_COPY)
std::unordered_set<std::string>& get_all_gpuarch();
#endif
inline
std::string transmogrify_triple(const std::string& triple)
{
@@ -43,7 +45,6 @@ std::string transmogrify_triple(const std::string& triple)
if (triple.find(old_prefix) == 0) {
return new_prefix + triple.substr(sizeof(old_prefix) - 1);
}
return (triple.find(new_prefix) == 0) ? triple : "";
}
@@ -114,9 +115,7 @@ class Bundled_code_header {
friend inline bool read(RandomAccessIterator f, RandomAccessIterator l,
Bundled_code_header& x) {
if (f == l) return false;
std::copy_n(f, sizeof(x.header_.cbuf_), x.header_.cbuf_);
if (valid(x)) {
x.bundles_.resize(x.header_.bundle_cnt_);
@@ -126,11 +125,16 @@ class Bundled_code_header {
it += sizeof(y.header.cbuf);
y.triple.assign(it, it + y.header.triple_sz);
#ifdef DISABLE_REDUCED_GPU_BLOB_COPY
std::copy_n(f + y.header.offset, y.header.bundle_sz, std::back_inserter(y.blob));
#else
auto& gpuArch = get_all_gpuarch();
auto itgpuArch = std::find(gpuArch.begin(),gpuArch.end(),y.triple);
if (itgpuArch != gpuArch.end()){
std::copy_n(f + y.header.offset, y.header.bundle_sz, std::back_inserter(y.blob));
}
#endif
it += y.header.triple_sz;
x.bundled_code_size = std::max(x.bundled_code_size,
y.header.offset + y.header.bundle_sz);
}
hipamd/include/hip/hcc_detail/device_functions.h
+134
View File
@@ -319,6 +319,41 @@ double __shfl(double var, int src_lane, int width = warpSize) {
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
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
@@ -356,6 +391,39 @@ double __shfl_up(double var, unsigned int lane_delta, int width = warpSize) {
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
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
@@ -393,6 +461,39 @@ double __shfl_down(double var, unsigned int lane_delta, int width = warpSize) {
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
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
@@ -430,6 +531,39 @@ double __shfl_xor(double var, int lane_mask, int width = warpSize) {
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_down(static_cast<int>(var), lane_delta, 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;
}
#define MASK1 0x00ff00ff
#define MASK2 0xff00ff00
hipamd/include/hip/hcc_detail/driver_types.h
+3 -6
View File
@@ -255,14 +255,14 @@ typedef struct hipMemcpy3DParms {
hipArray_t srcArray;
struct hipPos srcPos;
struct hipPitchedPtr srcPtr;
hipArray_t dstArray;
struct hipPos dstPos;
struct hipPitchedPtr dstPtr;
struct hipExtent extent;
enum hipMemcpyKind kind;
} hipMemcpy3DParms;
typedef struct HIP_MEMCPY3D {
size_t Depth;
size_t Height;
size_t WidthInBytes;
@@ -283,10 +283,7 @@ typedef struct hipMemcpy3DParms {
size_t srcLOD;
hipMemoryType srcMemoryType;
size_t srcPitch;
size_t srcXInBytes;
size_t srcY;
size_t srcZ;
}hipMemcpy3DParms;
} HIP_MEMCPY3D;
static inline struct hipPitchedPtr make_hipPitchedPtr(void* d, size_t p, size_t xsz,
size_t ysz) {
hipamd/include/hip/hcc_detail/hip_runtime_api.h
+32 -4
View File
@@ -212,6 +212,11 @@ enum hipLimit_t {
#define hipArrayCubemap 0x04
#define hipArrayTextureGather 0x08
#define hipOccupancyDefault 0x00
#define hipCooperativeLaunchMultiDeviceNoPreSync 0x01
#define hipCooperativeLaunchMultiDeviceNoPostSync 0x02
/*
* @brief hipJitOption
* @enum
@@ -2903,7 +2908,7 @@ hipError_t hipModuleLaunchKernel(hipFunction_t f, unsigned int gridDimX, unsigne
* @param [in] stream Stream where the kernel should be dispatched. May be 0, in which case th
* default stream is used with associated synchronization rules.
*
* @returns hipSuccess, hipInvalidDevice, hipErrorNotInitialized, hipErrorInvalidValue
* @returns hipSuccess, hipInvalidDevice, hipErrorNotInitialized, hipErrorInvalidValue, hipErrorCooperativeLaunchTooLarge
*/
hipError_t hipLaunchCooperativeKernel(const void* f, dim3 gridDim, dim3 blockDimX,
void** kernelParams, unsigned int sharedMemBytes,
@@ -2917,7 +2922,7 @@ hipError_t hipLaunchCooperativeKernel(const void* f, dim3 gridDim, dim3 blockDim
* @param [in] numDevices Size of the launchParamsList array.
* @param [in] flags Flags to control launch behavior.
*
* @returns hipSuccess, hipInvalidDevice, hipErrorNotInitialized, hipErrorInvalidValue
* @returns hipSuccess, hipInvalidDevice, hipErrorNotInitialized, hipErrorInvalidValue, hipErrorCooperativeLaunchTooLarge
*/
hipError_t hipLaunchCooperativeKernelMultiDevice(hipLaunchParams* launchParamsList,
int numDevices, unsigned int flags);
@@ -2954,13 +2959,36 @@ hipError_t hipOccupancyMaxActiveBlocksPerMultiprocessor(
* @brief Returns occupancy for a device function.
*
* @param [out] numBlocks Returned occupancy
* @param [in] func Kernel function for which occupancy is calulated
* @param [in] func Kernel function (hipFunction) 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
*/
hipError_t hipDrvOccupancyMaxActiveBlocksPerMultiprocessor(
int* numBlocks, hipFunction_t f, int blockSize, size_t dynSharedMemPerBlk);
/**
* @brief Returns occupancy for a device function.
*
* @param [out] numBlocks Returned occupancy
* @param [in] f 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] 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);
uint32_t* numBlocks, hipFunction_t f, uint32_t blockSize, size_t dynSharedMemPerBlk, unsigned int flags __dparm(hipOccupancyDefault));
/**
* @brief Returns occupancy for a device function.
*
* @param [out] numBlocks Returned occupancy
* @param [in] f Kernel function(hipFunction_t) 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] flags Extra flags for occupancy calculation (currently ignored)
*/
hipError_t hipDrvOccupancyMaxActiveBlocksPerMultiprocessorWithFlags(
int* numBlocks, hipFunction_t f, int blockSize, size_t dynSharedMemPerBlk, unsigned int flags);
/**
* @brief Launches kernels on multiple devices and guarantees all specified kernels are dispatched
hipamd/include/hip/hcc_detail/hiprtc.h
+20 -46
View File
@@ -19,10 +19,14 @@ 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
#ifndef HIPRTC_H
#define HIPRTC_H
#include <cstddef>
#include <string>
#ifdef __cplusplus
extern "C" {
#endif /* __cplusplus */
#include <stdlib.h>
enum hiprtcResult {
HIPRTC_SUCCESS = 0,
@@ -41,29 +45,22 @@ enum hiprtcResult {
const char* hiprtcGetErrorString(hiprtcResult result);
inline
hiprtcResult hiprtcVersion(int* major, int* minor) noexcept
{ // TODO: NVRTC versioning is somewhat unclear.
if (!major || !minor) return HIPRTC_ERROR_INVALID_INPUT;
// TODO: this should be generic / set by the build infrastructure.
*major = 9;
*minor = 0;
hiprtcResult hiprtcVersion(int* major, int* minor);
return HIPRTC_SUCCESS;
}
struct _hiprtcProgram;
using hiprtcProgram = _hiprtcProgram*;
typedef struct _hiprtcProgram* hiprtcProgram;
hiprtcResult hiprtcAddNameExpression(hiprtcProgram prog,
const char* name_expression);
hiprtcResult hiprtcCompileProgram(hiprtcProgram prog, int numOptions,
hiprtcResult hiprtcCompileProgram(hiprtcProgram prog,
int numOptions,
const char** options);
hiprtcResult hiprtcCreateProgram(hiprtcProgram* prog, const char* src,
const char* name, int numHeaders,
hiprtcResult hiprtcCreateProgram(hiprtcProgram* prog,
const char* src,
const char* name,
int numHeaders,
const char** headers,
const char** includeNames);
@@ -76,37 +73,14 @@ hiprtcResult hiprtcGetLoweredName(hiprtcProgram prog,
hiprtcResult hiprtcGetProgramLog(hiprtcProgram prog, char* log);
hiprtcResult hiprtcGetProgramLogSize(hiprtcProgram prog,
std::size_t* logSizeRet);
size_t* logSizeRet);
hiprtcResult hiprtcGetCode(hiprtcProgram prog, char* code);
hiprtcResult hiprtcGetCodeSize(hiprtcProgram prog, std::size_t* codeSizeRet);
hiprtcResult hiprtcGetCodeSize(hiprtcProgram prog, size_t* codeSizeRet);
namespace hip_impl
{
char* demangle(const char* mangled_expression);
#ifdef __cplusplus
}
#endif /* __cplusplus */
#if defined(HIPRTC_GET_TYPE_NAME)
#include <typeinfo>
#if defined(_WIN32)
#include <dbghelp.h>
template<typename>
hiprtcResult hiprtcGetTypeName(std::string*) = delete;
#else
template<typename T>
inline
hiprtcResult hiprtcGetTypeName(std::string* result)
{
if (!result) return HIPRTC_ERROR_INVALID_INPUT;
char * res= hip_impl::demangle(typeid(T).name());
result->assign(res == nullptr ? "" : res);
std::free(res);
return (result->empty()) ? HIPRTC_ERROR_INTERNAL_ERROR :
HIPRTC_SUCCESS;
}
#endif
#endif
#endif //HIPRTC_H
hipamd/include/hip/hip_profile.h
-18
View File
@@ -1,16 +1,13 @@
/*
Copyright (c) 2015 - 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
@@ -23,23 +20,8 @@ THE SOFTWARE.
#ifndef HIP_INCLUDE_HIP_HIP_PROFILE_H
#define HIP_INCLUDE_HIP_HIP_PROFILE_H
#warning "HIP Profiling through markers is deprecated, please check roctrace/rocTX support."
#if not defined(ENABLE_HIP_PROFILE)
#define ENABLE_HIP_PROFILE 1
#endif
#if defined(__HIP_PLATFORM_HCC__) and (ENABLE_HIP_PROFILE == 1)
#warning "HIP Markers are deprecated and would be removed soon."
#include <CXLActivityLogger.h>
#define HIP_SCOPED_MARKER(markerName, group) \
amdtScopedMarker __scopedMarker(markerName, group, nullptr);
#define HIP_BEGIN_MARKER(markerName, group) amdtBeginMarker(markerName, group, nullptr);
#define HIP_END_MARKER() amdtEndMarker();
#else
#define HIP_SCOPED_MARKER(markerName, group)
#define HIP_BEGIN_MARKER(markerName, group)
#define HIP_END_MARKER()
#endif
#endif
hipamd/include/hip/hip_runtime_api.h
+4
View File
@@ -255,6 +255,10 @@ typedef enum __HIP_NODISCARD hipError_t {
713, ///< Produced when trying to unlock a non-page-locked memory.
hipErrorLaunchFailure =
719, ///< An exception occurred on the device while executing a kernel.
hipErrorCooperativeLaunchTooLarge =
720, ///< This error indicates that the number of blocks launched per grid for a kernel
///< that was launched via cooperative launch APIs exceeds the maximum number of
///< allowed blocks for the current device
hipErrorNotSupported = 801, ///< Produced when the hip API is not supported/implemented
hipErrorUnknown = 999, //< Unknown error.
// HSA Runtime Error Codes start here.
hipamd/include/hip/nvcc_detail/hip_runtime_api.h
+14
View File
@@ -140,6 +140,14 @@ typedef enum cudaChannelFormatKind hipChannelFormatKind;
#define hipLimitMallocHeapSize cudaLimitMallocHeapSize
#define hipIpcMemLazyEnablePeerAccess cudaIpcMemLazyEnablePeerAccess
#define hipOccupancyDefault cudaOccupancyDefault
#define hipCooperativeLaunchMultiDeviceNoPreSync \
cudaCooperativeLaunchMultiDeviceNoPreSync
#define hipCooperativeLaunchMultiDeviceNoPostSync \
cudaCooperativeLaunchMultiDeviceNoPostSync
// enum CUjit_option redefines
#define hipJitOptionMaxRegisters CU_JIT_MAX_REGISTERS
#define hipJitOptionThreadsPerBlock CU_JIT_THREADS_PER_BLOCK
@@ -267,6 +275,8 @@ inline static hipError_t hipCUDAErrorTohipError(cudaError_t cuError) {
return hipErrorNotInitialized;
case cudaErrorLaunchFailure:
return hipErrorLaunchFailure;
case cudaErrorCooperativeLaunchTooLarge:
return hipErrorCooperativeLaunchTooLarge;
case cudaErrorPriorLaunchFailure:
return hipErrorPriorLaunchFailure;
case cudaErrorLaunchOutOfResources:
@@ -445,6 +455,8 @@ inline static hipError_t hipCUResultTohipError(CUresult cuError) {
return hipErrorHostMemoryNotRegistered;
case CUDA_ERROR_LAUNCH_FAILED:
return hipErrorLaunchFailure;
case CUDA_ERROR_COOPERATIVE_LAUNCH_TOO_LARGE:
return hipErrorCooperativeLaunchTooLarge;
case CUDA_ERROR_NOT_SUPPORTED:
return hipErrorNotSupported;
case CUDA_ERROR_UNKNOWN:
@@ -601,6 +613,8 @@ inline static cudaError_t hipErrorToCudaError(hipError_t hError) {
return cudaErrorSetOnActiveProcess;
case hipErrorLaunchFailure:
return cudaErrorLaunchFailure;
case hipErrorCooperativeLaunchTooLarge:
return cudaErrorCooperativeLaunchTooLarge;
case hipErrorNotSupported:
return cudaErrorNotSupported;
// HSA: does not exist in CUDA