Saleel Kudchadker
954c70ace8
- For H2D transfer cases, we cannot enfore ErrorNotReady behavior if the
source is unpinned. In that case the driver can sync depending on if
it pins or stages the source buffer.
Change-Id: I436a4b667bff172cfbadc58631a46a215c40fcb1
[ROCm/hip-tests commit: d8f5c2560f]
551 lines
24 KiB
C++
551 lines
24 KiB
C++
/*
|
|
Copyright (c) 2022 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
|
|
|
|
#include <variant>
|
|
|
|
#include <hip_test_common.hh>
|
|
#include <hip/hip_runtime_api.h>
|
|
#include <utils.hh>
|
|
#include <resource_guards.hh>
|
|
#include <hip/driver_types.h>
|
|
|
|
template <bool should_synchronize, bool unaligned = false, typename F>
|
|
void Memcpy2DDeviceToHostShell(F memcpy_func, const hipStream_t kernel_stream = nullptr) {
|
|
const auto kind = GENERATE(hipMemcpyDeviceToHost, hipMemcpyDefault);
|
|
|
|
constexpr size_t cols = 127;
|
|
constexpr size_t rows = 128;
|
|
|
|
LinearAllocGuard2D<int, unaligned> device_alloc(cols, rows);
|
|
|
|
const size_t host_pitch = GENERATE_REF(device_alloc.width(), device_alloc.width() + 64);
|
|
LinearAllocGuard<int> host_alloc(LinearAllocs::hipHostMalloc, host_pitch * rows);
|
|
|
|
const dim3 threads_per_block(32, 32);
|
|
const dim3 blocks(cols / threads_per_block.x + 1, rows / threads_per_block.y + 1);
|
|
Iota<<<blocks, threads_per_block>>>(device_alloc.ptr(), device_alloc.pitch(),
|
|
device_alloc.width_logical(), device_alloc.height(), 1);
|
|
HIP_CHECK(hipGetLastError());
|
|
HIP_CHECK(hipDeviceSynchronize());
|
|
|
|
HIP_CHECK(memcpy_func(host_alloc.ptr(), host_pitch, device_alloc.ptr(), device_alloc.pitch(),
|
|
device_alloc.width(), device_alloc.height(), kind));
|
|
if constexpr (should_synchronize) {
|
|
HIP_CHECK(hipStreamSynchronize(kernel_stream));
|
|
}
|
|
|
|
constexpr auto f = [](size_t x, size_t y, size_t z) { return z * cols * rows + y * cols + x; };
|
|
PitchedMemoryVerify(host_alloc.ptr(), host_pitch, device_alloc.width_logical(),
|
|
device_alloc.height(), 1, f);
|
|
}
|
|
|
|
template <bool should_synchronize, bool enable_peer_access, bool unaligned = false, typename F>
|
|
void Memcpy2DDeviceToDeviceShell(F memcpy_func, const hipStream_t kernel_stream = nullptr) {
|
|
const auto kind = GENERATE(hipMemcpyDeviceToDevice, hipMemcpyDefault);
|
|
|
|
constexpr size_t cols = 127;
|
|
constexpr size_t rows = 128;
|
|
|
|
const auto device_count = HipTest::getDeviceCount();
|
|
const auto src_device = GENERATE_COPY(range(0, device_count));
|
|
const auto dst_device = GENERATE_COPY(range(0, device_count));
|
|
const size_t src_cols_mult = GENERATE(1, 2);
|
|
|
|
INFO("Src device: " << src_device << ", Dst device: " << dst_device);
|
|
if (device_count > 1) {
|
|
int can_access_peer = 0;
|
|
HIP_CHECK(hipDeviceCanAccessPeer(&can_access_peer, src_device, dst_device));
|
|
if (!can_access_peer) {
|
|
std::string msg = "Skipped as peer access cannot be enabled between devices " +
|
|
std::to_string(src_device) + " " + std::to_string(dst_device);
|
|
HipTest::HIP_SKIP_TEST(msg.c_str());
|
|
return;
|
|
}
|
|
}
|
|
|
|
HIP_CHECK(hipSetDevice(src_device));
|
|
if constexpr (enable_peer_access) {
|
|
if (src_device == dst_device) {
|
|
return;
|
|
}
|
|
HIP_CHECK(hipDeviceEnablePeerAccess(dst_device, 0));
|
|
}
|
|
|
|
LinearAllocGuard2D<int, unaligned> src_alloc(cols * src_cols_mult, rows);
|
|
HIP_CHECK(hipSetDevice(dst_device));
|
|
LinearAllocGuard2D<int, unaligned> dst_alloc(cols, rows);
|
|
HIP_CHECK(hipSetDevice(src_device));
|
|
LinearAllocGuard<int> host_alloc(LinearAllocs::hipHostMalloc, dst_alloc.width() * rows);
|
|
|
|
const dim3 threads_per_block(32, 32);
|
|
const dim3 blocks(cols / threads_per_block.x + 1, rows / threads_per_block.y + 1);
|
|
// Using dst_alloc width and height to set only the elements that will be copied over to
|
|
// dst_alloc
|
|
Iota<<<blocks, threads_per_block>>>(src_alloc.ptr(), src_alloc.pitch(), dst_alloc.width_logical(),
|
|
dst_alloc.height(), 1);
|
|
HIP_CHECK(hipGetLastError());
|
|
HIP_CHECK(hipDeviceSynchronize());
|
|
|
|
HIP_CHECK(memcpy_func(dst_alloc.ptr(), dst_alloc.pitch(), src_alloc.ptr(), src_alloc.pitch(),
|
|
dst_alloc.width(), dst_alloc.height(), kind));
|
|
if constexpr (should_synchronize) {
|
|
HIP_CHECK(hipStreamSynchronize(kernel_stream));
|
|
}
|
|
|
|
HIP_CHECK(hipMemcpy2D(host_alloc.ptr(), dst_alloc.width(), dst_alloc.ptr(), dst_alloc.pitch(),
|
|
dst_alloc.width(), dst_alloc.height(), hipMemcpyDeviceToHost));
|
|
constexpr auto f = [](size_t x, size_t y, size_t z) { return z * cols * rows + y * cols + x; };
|
|
PitchedMemoryVerify(host_alloc.ptr(), dst_alloc.width(), dst_alloc.width_logical(),
|
|
dst_alloc.height(), 1, f);
|
|
}
|
|
|
|
template <bool should_synchronize, bool unaligned = false, typename F>
|
|
void Memcpy2DHostToDeviceShell(F memcpy_func, const hipStream_t kernel_stream = nullptr) {
|
|
const auto kind = GENERATE(hipMemcpyHostToDevice, hipMemcpyDefault);
|
|
|
|
constexpr size_t cols = 127;
|
|
constexpr size_t rows = 128;
|
|
|
|
LinearAllocGuard2D<int, unaligned> device_alloc(cols, rows);
|
|
|
|
const size_t host_pitch = GENERATE_REF(device_alloc.pitch(), 2 * device_alloc.pitch());
|
|
|
|
LinearAllocGuard<int> src_host_alloc(LinearAllocs::hipHostMalloc, host_pitch * rows);
|
|
LinearAllocGuard<int> dst_host_alloc(LinearAllocs::hipHostMalloc, device_alloc.width() * rows);
|
|
|
|
constexpr auto f = [](size_t x, size_t y, size_t z) { return z * cols * rows + y * cols + x; };
|
|
PitchedMemorySet(src_host_alloc.ptr(), host_pitch, device_alloc.width_logical(),
|
|
device_alloc.height(), 1, f);
|
|
|
|
std::fill_n(dst_host_alloc.ptr(), device_alloc.width_logical() * rows, 0);
|
|
|
|
HIP_CHECK(memcpy_func(device_alloc.ptr(), device_alloc.pitch(), src_host_alloc.ptr(), host_pitch,
|
|
device_alloc.width(), device_alloc.height(), kind));
|
|
if constexpr (should_synchronize) {
|
|
HIP_CHECK(hipStreamSynchronize(kernel_stream));
|
|
}
|
|
|
|
HIP_CHECK(hipMemcpy2D(dst_host_alloc.ptr(), device_alloc.width(), device_alloc.ptr(),
|
|
device_alloc.pitch(), device_alloc.width(), device_alloc.height(),
|
|
hipMemcpyDeviceToHost));
|
|
|
|
PitchedMemoryVerify(dst_host_alloc.ptr(), device_alloc.width(), device_alloc.width_logical(),
|
|
device_alloc.height(), 1, f);
|
|
}
|
|
|
|
template <bool should_synchronize, typename F>
|
|
void Memcpy2DHostToHostShell(F memcpy_func, const hipStream_t kernel_stream = nullptr) {
|
|
const auto kind = GENERATE(hipMemcpyHostToHost, hipMemcpyDefault);
|
|
|
|
constexpr size_t cols = 127;
|
|
constexpr size_t rows = 128;
|
|
|
|
const size_t src_pitch = GENERATE_REF(cols * sizeof(int), cols * sizeof(int) + 64);
|
|
|
|
LinearAllocGuard<int> src_host(LinearAllocs::hipHostMalloc, src_pitch * rows);
|
|
LinearAllocGuard<int> dst_host(LinearAllocs::hipHostMalloc, cols * sizeof(int) * rows);
|
|
|
|
constexpr auto f = [](size_t x, size_t y, size_t z) { return z * cols * rows + y * cols + x; };
|
|
PitchedMemorySet(src_host.ptr(), src_pitch, cols, rows, 1, f);
|
|
|
|
HIP_CHECK(memcpy_func(dst_host.ptr(), cols * sizeof(int), src_host.ptr(), src_pitch,
|
|
cols * sizeof(int), rows, kind));
|
|
if constexpr (should_synchronize) {
|
|
HIP_CHECK(hipStreamSynchronize(kernel_stream));
|
|
}
|
|
|
|
PitchedMemoryVerify(dst_host.ptr(), cols * sizeof(int), cols, rows, 1, f);
|
|
}
|
|
|
|
// Synchronization behavior checks
|
|
template <typename F>
|
|
void MemcpySyncBehaviorCheck(F memcpy_func, const bool should_sync,
|
|
const hipStream_t kernel_stream) {
|
|
LaunchDelayKernel(std::chrono::milliseconds{300}, kernel_stream);
|
|
HIP_CHECK(memcpy_func());
|
|
if (should_sync) {
|
|
HIP_CHECK(hipStreamSynchronize(kernel_stream));
|
|
HIP_CHECK(hipStreamQuery(kernel_stream));
|
|
} else {
|
|
HIP_CHECK_ERROR(hipStreamQuery(kernel_stream), hipErrorNotReady);
|
|
}
|
|
}
|
|
|
|
template <bool unaligned = false, typename F>
|
|
void Memcpy2DHtoDSyncBehavior(F memcpy_func, const bool should_sync,
|
|
const hipStream_t kernel_stream = nullptr) {
|
|
using LA = LinearAllocs;
|
|
LinearAllocGuard<int> host_alloc(LA::hipHostMalloc, 32 * sizeof(int) * 32);
|
|
LinearAllocGuard2D<int, unaligned> device_alloc(32, 32);
|
|
MemcpySyncBehaviorCheck(std::bind(memcpy_func, device_alloc.ptr(), device_alloc.pitch(),
|
|
host_alloc.ptr(), device_alloc.width(), device_alloc.width(),
|
|
device_alloc.height(), hipMemcpyHostToDevice),
|
|
should_sync, kernel_stream);
|
|
}
|
|
|
|
template <bool unaligned = false, typename F>
|
|
void Memcpy2DDtoHPageableSyncBehavior(F memcpy_func, const bool should_sync,
|
|
const hipStream_t kernel_stream = nullptr) {
|
|
LinearAllocGuard<int> host_alloc(LinearAllocs::malloc, 32 * sizeof(int) * 32);
|
|
LinearAllocGuard2D<int, unaligned> device_alloc(32, 32);
|
|
MemcpySyncBehaviorCheck(std::bind(memcpy_func, host_alloc.ptr(), device_alloc.width(),
|
|
device_alloc.ptr(), device_alloc.pitch(), device_alloc.width(),
|
|
device_alloc.height(), hipMemcpyDeviceToHost),
|
|
should_sync, kernel_stream);
|
|
}
|
|
|
|
template <bool unaligned = false, typename F>
|
|
void Memcpy2DDtoHPinnedSyncBehavior(F memcpy_func, const bool should_sync,
|
|
const hipStream_t kernel_stream = nullptr) {
|
|
LinearAllocGuard<int> host_alloc(LinearAllocs::hipHostMalloc, 32 * sizeof(int) * 32);
|
|
LinearAllocGuard2D<int, unaligned> device_alloc(32, 32);
|
|
MemcpySyncBehaviorCheck(std::bind(memcpy_func, host_alloc.ptr(), device_alloc.width(),
|
|
device_alloc.ptr(), device_alloc.pitch(), device_alloc.width(),
|
|
device_alloc.height(), hipMemcpyDeviceToHost),
|
|
should_sync, kernel_stream);
|
|
}
|
|
|
|
template <bool unaligned = false, typename F>
|
|
void Memcpy2DDtoDSyncBehavior(F memcpy_func, const bool should_sync,
|
|
const hipStream_t kernel_stream = nullptr) {
|
|
LinearAllocGuard2D<int, unaligned> src_alloc(32, 32);
|
|
LinearAllocGuard2D<int, unaligned> dst_alloc(32, 32);
|
|
MemcpySyncBehaviorCheck(
|
|
std::bind(memcpy_func, dst_alloc.ptr(), dst_alloc.pitch(), src_alloc.ptr(), src_alloc.pitch(),
|
|
dst_alloc.width(), dst_alloc.height(), hipMemcpyDeviceToDevice),
|
|
should_sync, kernel_stream);
|
|
}
|
|
|
|
template <typename F>
|
|
void Memcpy2DHtoHSyncBehavior(F memcpy_func, const bool should_sync,
|
|
const hipStream_t kernel_stream = nullptr) {
|
|
using LA = LinearAllocs;
|
|
const auto src_alloc_type = GENERATE(LA::malloc, LA::hipHostMalloc);
|
|
const auto dst_alloc_type = GENERATE(LA::malloc, LA::hipHostMalloc);
|
|
|
|
LinearAllocGuard<int> src_alloc(src_alloc_type, 32 * sizeof(int) * 32);
|
|
LinearAllocGuard<int> dst_alloc(dst_alloc_type, 32 * sizeof(int) * 32);
|
|
MemcpySyncBehaviorCheck(std::bind(memcpy_func, dst_alloc.ptr(), 32 * sizeof(int), src_alloc.ptr(),
|
|
32 * sizeof(int), 32 * sizeof(int), 32, hipMemcpyHostToHost),
|
|
should_sync, kernel_stream);
|
|
}
|
|
|
|
template <bool should_synchronize, bool unaligned = false, typename F>
|
|
void Memcpy2DZeroWidthHeight(F memcpy_func, const hipStream_t stream = nullptr) {
|
|
constexpr size_t cols = 63;
|
|
constexpr size_t rows = 64;
|
|
|
|
const auto [width_mult, height_mult] =
|
|
GENERATE(std::make_pair(0, 1), std::make_pair(1, 0), std::make_pair(0, 0));
|
|
|
|
SECTION("Device to Host") {
|
|
LinearAllocGuard2D<uint8_t, unaligned> device_alloc(cols, rows);
|
|
LinearAllocGuard<uint8_t> host_alloc(LinearAllocs::hipHostMalloc, device_alloc.width() * rows);
|
|
std::fill_n(host_alloc.ptr(), device_alloc.width_logical() * device_alloc.height(), 42);
|
|
HIP_CHECK(hipMemset2D(device_alloc.ptr(), device_alloc.pitch(), 1, device_alloc.width(),
|
|
device_alloc.height()));
|
|
|
|
HIP_CHECK(memcpy_func(host_alloc.ptr(), device_alloc.width(), device_alloc.ptr(),
|
|
device_alloc.pitch(), device_alloc.width() * width_mult,
|
|
device_alloc.height() * height_mult, hipMemcpyDeviceToHost));
|
|
if constexpr (should_synchronize) {
|
|
HIP_CHECK(hipStreamSynchronize(stream));
|
|
}
|
|
ArrayFindIfNot(host_alloc.ptr(), static_cast<uint8_t>(42),
|
|
device_alloc.width_logical() * device_alloc.height());
|
|
}
|
|
|
|
SECTION("Device to Device") {
|
|
LinearAllocGuard2D<uint8_t, unaligned> src_alloc(cols, rows);
|
|
LinearAllocGuard2D<uint8_t, unaligned> dst_alloc(cols, rows);
|
|
LinearAllocGuard<uint8_t> host_alloc(LinearAllocs::hipHostMalloc, dst_alloc.width() * rows);
|
|
HIP_CHECK(
|
|
hipMemset2D(src_alloc.ptr(), src_alloc.pitch(), 1, src_alloc.width(), src_alloc.height()));
|
|
HIP_CHECK(
|
|
hipMemset2D(dst_alloc.ptr(), dst_alloc.pitch(), 42, dst_alloc.width(), dst_alloc.height()));
|
|
HIP_CHECK(memcpy_func(dst_alloc.ptr(), dst_alloc.pitch(), src_alloc.ptr(), src_alloc.pitch(),
|
|
dst_alloc.width() * width_mult, dst_alloc.height() * height_mult,
|
|
hipMemcpyDeviceToDevice));
|
|
if constexpr (should_synchronize) {
|
|
HIP_CHECK(hipStreamSynchronize(stream));
|
|
}
|
|
HIP_CHECK(hipMemcpy2D(host_alloc.ptr(), dst_alloc.width(), dst_alloc.ptr(), dst_alloc.pitch(),
|
|
dst_alloc.width(), dst_alloc.height(), hipMemcpyDeviceToHost));
|
|
ArrayFindIfNot(host_alloc.ptr(), static_cast<uint8_t>(42),
|
|
dst_alloc.width_logical() * dst_alloc.height());
|
|
}
|
|
|
|
SECTION("Host to Device") {
|
|
LinearAllocGuard2D<uint8_t, unaligned> device_alloc(cols, rows);
|
|
LinearAllocGuard<uint8_t> src_host_alloc(LinearAllocs::hipHostMalloc,
|
|
device_alloc.width() * rows);
|
|
LinearAllocGuard<uint8_t> dst_host_alloc(LinearAllocs::hipHostMalloc,
|
|
device_alloc.width() * rows);
|
|
std::fill_n(src_host_alloc.ptr(), device_alloc.width_logical() * device_alloc.height(), 1);
|
|
HIP_CHECK(hipMemset2D(device_alloc.ptr(), device_alloc.pitch(), 42, device_alloc.width(),
|
|
device_alloc.height()));
|
|
HIP_CHECK(memcpy_func(device_alloc.ptr(), device_alloc.pitch(), src_host_alloc.ptr(),
|
|
device_alloc.width(), device_alloc.width() * width_mult,
|
|
device_alloc.height() * height_mult, hipMemcpyHostToDevice));
|
|
if constexpr (should_synchronize) {
|
|
HIP_CHECK(hipStreamSynchronize(stream));
|
|
}
|
|
HIP_CHECK(hipMemcpy2D(dst_host_alloc.ptr(), device_alloc.width(), device_alloc.ptr(),
|
|
device_alloc.pitch(), device_alloc.width(), device_alloc.height(),
|
|
hipMemcpyDeviceToHost));
|
|
ArrayFindIfNot(dst_host_alloc.ptr(), static_cast<uint8_t>(42),
|
|
device_alloc.width_logical() * device_alloc.height());
|
|
}
|
|
|
|
SECTION("Host to Host") {
|
|
const auto alloc_size = cols * rows;
|
|
LinearAllocGuard<uint8_t> src_alloc(LinearAllocs::hipHostMalloc, alloc_size);
|
|
LinearAllocGuard<uint8_t> dst_alloc(LinearAllocs::hipHostMalloc, alloc_size);
|
|
std::fill_n(src_alloc.ptr(), alloc_size, 1);
|
|
std::fill_n(dst_alloc.ptr(), alloc_size, 42);
|
|
HIP_CHECK(memcpy_func(dst_alloc.ptr(), cols, src_alloc.ptr(), cols, cols * width_mult,
|
|
rows * height_mult, hipMemcpyHostToHost));
|
|
if constexpr (should_synchronize) {
|
|
HIP_CHECK(hipStreamSynchronize(stream));
|
|
}
|
|
ArrayFindIfNot(dst_alloc.ptr(), static_cast<uint8_t>(42), alloc_size);
|
|
}
|
|
}
|
|
|
|
constexpr auto MemTypeHost() {
|
|
return hipMemoryTypeHost;
|
|
}
|
|
|
|
constexpr auto MemTypeDevice() {
|
|
return hipMemoryTypeDevice;
|
|
}
|
|
|
|
constexpr auto MemTypeArray() {
|
|
return hipMemoryTypeArray;
|
|
}
|
|
|
|
constexpr auto MemTypeUnified() {
|
|
return hipMemoryTypeUnified;
|
|
}
|
|
|
|
using PtrVariant = std::variant<void*, hipArray_t>;
|
|
|
|
constexpr void InitializeMemcpy2DParams(hip_Memcpy2D* parms, PtrVariant dst, size_t dpitch,
|
|
PtrVariant src, size_t spitch, size_t width, size_t height,
|
|
hipMemcpyKind kind, hipExtent src_offset = {0, 0, 0},
|
|
hipExtent dst_offset = {0, 0, 0}) {
|
|
if (std::holds_alternative<hipArray_t>(dst)) {
|
|
parms->dstMemoryType = MemTypeArray();
|
|
parms->dstArray = std::get<hipArray_t>(dst);
|
|
} else {
|
|
parms->dstPitch = dpitch;
|
|
auto ptr = std::get<void*>(dst);
|
|
switch (kind) {
|
|
case hipMemcpyDeviceToHost:
|
|
case hipMemcpyHostToHost:
|
|
parms->dstMemoryType = MemTypeHost();
|
|
parms->dstHost = ptr;
|
|
break;
|
|
case hipMemcpyDeviceToDevice:
|
|
case hipMemcpyHostToDevice:
|
|
parms->dstMemoryType = MemTypeDevice();
|
|
parms->dstDevice = reinterpret_cast<hipDeviceptr_t>(ptr);
|
|
break;
|
|
case hipMemcpyDefault:
|
|
parms->dstMemoryType = MemTypeUnified();
|
|
parms->dstDevice = reinterpret_cast<hipDeviceptr_t>(ptr);
|
|
break;
|
|
default:
|
|
assert(false);
|
|
}
|
|
}
|
|
|
|
if (std::holds_alternative<hipArray_t>(src)) {
|
|
parms->srcMemoryType = MemTypeArray();
|
|
parms->srcArray = std::get<hipArray_t>(src);
|
|
} else {
|
|
parms->srcPitch = spitch;
|
|
auto ptr = std::get<void*>(src);
|
|
switch (kind) {
|
|
case hipMemcpyDeviceToHost:
|
|
case hipMemcpyDeviceToDevice:
|
|
parms->srcMemoryType = MemTypeDevice();
|
|
parms->srcDevice = reinterpret_cast<hipDeviceptr_t>(ptr);
|
|
break;
|
|
case hipMemcpyHostToDevice:
|
|
case hipMemcpyHostToHost:
|
|
parms->srcMemoryType = MemTypeHost();
|
|
parms->srcHost = ptr;
|
|
break;
|
|
case hipMemcpyDefault:
|
|
parms->srcMemoryType = MemTypeUnified();
|
|
parms->srcDevice = reinterpret_cast<hipDeviceptr_t>(ptr);
|
|
break;
|
|
default:
|
|
assert(false);
|
|
}
|
|
}
|
|
|
|
parms->WidthInBytes = width;
|
|
parms->Height = height;
|
|
parms->srcXInBytes = src_offset.width;
|
|
parms->srcY = src_offset.height;
|
|
parms->dstXInBytes = dst_offset.width;
|
|
parms->dstY = dst_offset.height;
|
|
}
|
|
|
|
template <bool async = false>
|
|
constexpr auto MemcpyParam2DAdapter(const hipExtent src_offset = {0, 0, 0},
|
|
const hipExtent dst_offset = {0, 0, 0}) {
|
|
return [=](PtrVariant dst, size_t dpitch, PtrVariant src, size_t spitch, size_t width,
|
|
size_t height, hipMemcpyKind kind, hipStream_t stream = nullptr) {
|
|
hip_Memcpy2D parms = {};
|
|
memset(&parms, 0x0, sizeof(hip_Memcpy2D));
|
|
|
|
InitializeMemcpy2DParams(&parms, dst, dpitch, src, spitch, width, height, kind, src_offset,
|
|
dst_offset);
|
|
|
|
if constexpr (async) {
|
|
return hipMemcpyParam2DAsync(&parms, stream);
|
|
} else {
|
|
return hipMemcpyParam2D(&parms);
|
|
}
|
|
};
|
|
}
|
|
|
|
#if HT_AMD
|
|
constexpr auto DrvMemcpy2DUnalignedAdapter() {
|
|
return [=](PtrVariant dst, size_t dpitch, PtrVariant src, size_t spitch, size_t width,
|
|
size_t height, hipMemcpyKind kind, hipStream_t stream = nullptr) {
|
|
(void)stream;
|
|
hip_Memcpy2D parms = {};
|
|
memset(&parms, 0x0, sizeof(hip_Memcpy2D));
|
|
|
|
InitializeMemcpy2DParams(&parms, dst, dpitch, src, spitch, width, height, kind);
|
|
return hipDrvMemcpy2DUnaligned(&parms);
|
|
};
|
|
}
|
|
#endif
|
|
|
|
template <bool should_synchronize, typename F>
|
|
void MemcpyParam2DArrayHostShell(F memcpy_func, const hipStream_t kernel_stream = nullptr) {
|
|
hipExtent extent{127 * sizeof(int), 128, 1};
|
|
|
|
LinearAllocGuard<int> src_host(LinearAllocs::hipHostMalloc,
|
|
extent.width * extent.height * extent.depth);
|
|
LinearAllocGuard<int> dst_host(LinearAllocs::hipHostMalloc,
|
|
extent.width * extent.height * extent.depth);
|
|
|
|
DrvArrayAllocGuard<int> src_array(extent);
|
|
DrvArrayAllocGuard<int> dst_array(extent);
|
|
|
|
const auto f = [extent](size_t x, size_t y, size_t z) {
|
|
auto width_logical = extent.width / sizeof(int);
|
|
return z * width_logical * extent.height + y * width_logical + x;
|
|
};
|
|
PitchedMemorySet(src_host.ptr(), extent.width, extent.width / sizeof(int), extent.height,
|
|
extent.depth, f);
|
|
|
|
// Host -> Array
|
|
HIP_CHECK(memcpy_func(src_array.ptr(), 0, src_host.ptr(), extent.width, extent.width,
|
|
extent.height, hipMemcpyHostToDevice, kernel_stream));
|
|
if constexpr (should_synchronize) {
|
|
HIP_CHECK(hipStreamSynchronize(kernel_stream));
|
|
}
|
|
|
|
// Array -> Array
|
|
HIP_CHECK(memcpy_func(dst_array.ptr(), 0, src_array.ptr(), 0, extent.width, extent.height,
|
|
hipMemcpyDeviceToDevice, kernel_stream));
|
|
if constexpr (should_synchronize) {
|
|
HIP_CHECK(hipStreamSynchronize(kernel_stream));
|
|
}
|
|
|
|
// Array -> Host
|
|
HIP_CHECK(memcpy_func(dst_host.ptr(), extent.width, dst_array.ptr(), 0, extent.width,
|
|
extent.height, hipMemcpyDeviceToHost, kernel_stream));
|
|
if constexpr (should_synchronize) {
|
|
HIP_CHECK(hipStreamSynchronize(kernel_stream));
|
|
}
|
|
|
|
PitchedMemoryVerify(dst_host.ptr(), extent.width, extent.width / sizeof(int), extent.height,
|
|
extent.depth, f);
|
|
}
|
|
|
|
template <bool should_synchronize, typename F>
|
|
void MemcpyParam2DArrayDeviceShell(F memcpy_func, const hipStream_t kernel_stream = nullptr) {
|
|
hipExtent extent{127 * sizeof(int), 128, 1};
|
|
|
|
LinearAllocGuard<int> host_alloc(LinearAllocs::hipHostMalloc,
|
|
extent.width * extent.height * extent.depth);
|
|
|
|
DrvArrayAllocGuard<int> src_array(extent);
|
|
DrvArrayAllocGuard<int> dst_array(extent);
|
|
|
|
LinearAllocGuard3D<int> src_device(extent);
|
|
LinearAllocGuard3D<int> dst_device(extent);
|
|
|
|
const dim3 threads_per_block(32, 32);
|
|
const dim3 blocks(src_device.width_logical() / threads_per_block.x + 1,
|
|
src_device.height() / threads_per_block.y + 1, src_device.depth());
|
|
Iota<<<blocks, threads_per_block>>>(src_device.ptr(), src_device.pitch(),
|
|
src_device.width_logical(), src_device.height(),
|
|
src_device.depth());
|
|
HIP_CHECK(hipGetLastError());
|
|
HIP_CHECK(hipDeviceSynchronize());
|
|
|
|
// Device -> Array
|
|
HIP_CHECK(memcpy_func(src_array.ptr(), 0, src_device.ptr(), src_device.pitch(), extent.width,
|
|
extent.height, hipMemcpyDeviceToDevice, kernel_stream));
|
|
if constexpr (should_synchronize) {
|
|
HIP_CHECK(hipStreamSynchronize(kernel_stream));
|
|
}
|
|
|
|
// Array -> Array
|
|
HIP_CHECK(memcpy_func(dst_array.ptr(), 0, src_array.ptr(), 0, extent.width, extent.height,
|
|
hipMemcpyDeviceToDevice, kernel_stream));
|
|
if constexpr (should_synchronize) {
|
|
HIP_CHECK(hipStreamSynchronize(kernel_stream));
|
|
}
|
|
|
|
// Array -> Device
|
|
HIP_CHECK(memcpy_func(dst_device.ptr(), dst_device.pitch(), dst_array.ptr(), 0, extent.width,
|
|
extent.height, hipMemcpyDeviceToDevice, kernel_stream));
|
|
if constexpr (should_synchronize) {
|
|
HIP_CHECK(hipStreamSynchronize(kernel_stream));
|
|
}
|
|
|
|
HIP_CHECK(memcpy_func(host_alloc.ptr(), extent.width, dst_device.ptr(), dst_device.pitch(),
|
|
extent.width, extent.height, hipMemcpyDeviceToHost, kernel_stream));
|
|
if constexpr (should_synchronize) {
|
|
HIP_CHECK(hipStreamSynchronize(kernel_stream));
|
|
}
|
|
|
|
const auto f = [extent](size_t x, size_t y, size_t z) {
|
|
auto width_logical = extent.width / sizeof(int);
|
|
return z * width_logical * extent.height + y * width_logical + x;
|
|
};
|
|
PitchedMemoryVerify(host_alloc.ptr(), extent.width, extent.width / sizeof(int), extent.height,
|
|
extent.depth, f);
|
|
}
|