Store AHWB in NN canonical memory type -- hal
Prior to this CL, the canonical memory type only held a SharedHandle,
which mirrors the behavior of native_handle_t/hidl_handle. This means
memory types including AHardwareBuffer were stored as this SharedHandle
type. With this CL, the canonical memory type is stored directly as
AHardwareBuffer to avoid using non-NDK AHardwareBuffer calls in the NN
runtime.
Bug: 179906132
Test: mma
Test: NeuralNetworksTest_static
Test: presubmit
Change-Id: I394071c193d15ac0c90ac47e5a2a9a79c635db6c
Merged-In: I394071c193d15ac0c90ac47e5a2a9a79c635db6c
(cherry picked from commit bbe43d950e981cfb5c06622c8f80b57ab60b0497)
diff --git a/neuralnetworks/aidl/utils/src/Conversions.cpp b/neuralnetworks/aidl/utils/src/Conversions.cpp
index 486d01b..db3504b 100644
--- a/neuralnetworks/aidl/utils/src/Conversions.cpp
+++ b/neuralnetworks/aidl/utils/src/Conversions.cpp
@@ -18,6 +18,8 @@
#include <aidl/android/hardware/common/NativeHandle.h>
#include <android-base/logging.h>
+#include <android/hardware_buffer.h>
+#include <cutils/native_handle.h>
#include <nnapi/OperandTypes.h>
#include <nnapi/OperationTypes.h>
#include <nnapi/Result.h>
@@ -27,6 +29,7 @@
#include <nnapi/Validation.h>
#include <nnapi/hal/CommonUtils.h>
#include <nnapi/hal/HandleError.h>
+#include <vndk/hardware_buffer.h>
#include <algorithm>
#include <chrono>
@@ -127,6 +130,61 @@
return canonical;
}
+GeneralResult<Handle> unvalidatedConvertHelper(const NativeHandle& aidlNativeHandle) {
+ std::vector<base::unique_fd> fds;
+ fds.reserve(aidlNativeHandle.fds.size());
+ for (const auto& fd : aidlNativeHandle.fds) {
+ const int dupFd = dup(fd.get());
+ if (dupFd == -1) {
+ // TODO(b/120417090): is ANEURALNETWORKS_UNEXPECTED_NULL the correct error to return
+ // here?
+ return NN_ERROR() << "Failed to dup the fd";
+ }
+ fds.emplace_back(dupFd);
+ }
+
+ return Handle{.fds = std::move(fds), .ints = aidlNativeHandle.ints};
+}
+
+struct NativeHandleDeleter {
+ void operator()(native_handle_t* handle) const {
+ if (handle) {
+ native_handle_close(handle);
+ native_handle_delete(handle);
+ }
+ }
+};
+
+using UniqueNativeHandle = std::unique_ptr<native_handle_t, NativeHandleDeleter>;
+
+static nn::GeneralResult<UniqueNativeHandle> nativeHandleFromAidlHandle(
+ const NativeHandle& handle) {
+ std::vector<base::unique_fd> fds;
+ fds.reserve(handle.fds.size());
+ for (const auto& fd : handle.fds) {
+ const int dupFd = dup(fd.get());
+ if (dupFd == -1) {
+ return NN_ERROR() << "Failed to dup the fd";
+ }
+ fds.emplace_back(dupFd);
+ }
+
+ constexpr size_t kIntMax = std::numeric_limits<int>::max();
+ CHECK_LE(handle.fds.size(), kIntMax);
+ CHECK_LE(handle.ints.size(), kIntMax);
+ native_handle_t* nativeHandle = native_handle_create(static_cast<int>(handle.fds.size()),
+ static_cast<int>(handle.ints.size()));
+ if (nativeHandle == nullptr) {
+ return NN_ERROR() << "Failed to create native_handle";
+ }
+ for (size_t i = 0; i < fds.size(); ++i) {
+ nativeHandle->data[i] = fds[i].release();
+ }
+ std::copy(handle.ints.begin(), handle.ints.end(), &nativeHandle->data[nativeHandle->numFds]);
+
+ return UniqueNativeHandle(nativeHandle);
+}
+
} // anonymous namespace
GeneralResult<OperandType> unvalidatedConvert(const aidl_hal::OperandType& operandType) {
@@ -318,10 +376,64 @@
return measureTiming ? MeasureTiming::YES : MeasureTiming::NO;
}
+static uint32_t roundUpToMultiple(uint32_t value, uint32_t multiple) {
+ return (value + multiple - 1) / multiple * multiple;
+}
+
GeneralResult<SharedMemory> unvalidatedConvert(const aidl_hal::Memory& memory) {
VERIFY_NON_NEGATIVE(memory.size) << "Memory size must not be negative";
+ if (memory.size > std::numeric_limits<uint32_t>::max()) {
+ return NN_ERROR() << "Memory: size must be <= std::numeric_limits<size_t>::max()";
+ }
+
+ if (memory.name != "hardware_buffer_blob") {
+ return std::make_shared<const Memory>(Memory{
+ .handle = NN_TRY(unvalidatedConvertHelper(memory.handle)),
+ .size = static_cast<uint32_t>(memory.size),
+ .name = memory.name,
+ });
+ }
+
+ const auto size = static_cast<uint32_t>(memory.size);
+ const auto format = AHARDWAREBUFFER_FORMAT_BLOB;
+ const auto usage = AHARDWAREBUFFER_USAGE_CPU_READ_OFTEN | AHARDWAREBUFFER_USAGE_CPU_WRITE_OFTEN;
+ const uint32_t width = size;
+ const uint32_t height = 1; // height is always 1 for BLOB mode AHardwareBuffer.
+ const uint32_t layers = 1; // layers is always 1 for BLOB mode AHardwareBuffer.
+
+ const UniqueNativeHandle handle = NN_TRY(nativeHandleFromAidlHandle(memory.handle));
+ const native_handle_t* nativeHandle = handle.get();
+
+ // AHardwareBuffer_createFromHandle() might fail because an allocator
+ // expects a specific stride value. In that case, we try to guess it by
+ // aligning the width to small powers of 2.
+ // TODO(b/174120849): Avoid stride assumptions.
+ AHardwareBuffer* hardwareBuffer = nullptr;
+ status_t status = UNKNOWN_ERROR;
+ for (uint32_t alignment : {1, 4, 32, 64, 128, 2, 8, 16}) {
+ const uint32_t stride = roundUpToMultiple(width, alignment);
+ AHardwareBuffer_Desc desc{
+ .width = width,
+ .height = height,
+ .layers = layers,
+ .format = format,
+ .usage = usage,
+ .stride = stride,
+ };
+ status = AHardwareBuffer_createFromHandle(&desc, nativeHandle,
+ AHARDWAREBUFFER_CREATE_FROM_HANDLE_METHOD_CLONE,
+ &hardwareBuffer);
+ if (status == NO_ERROR) {
+ break;
+ }
+ }
+ if (status != NO_ERROR) {
+ return NN_ERROR(ErrorStatus::GENERAL_FAILURE)
+ << "Can't create AHardwareBuffer from handle. Error: " << status;
+ }
+
return std::make_shared<const Memory>(Memory{
- .handle = NN_TRY(unvalidatedConvert(memory.handle)),
+ .handle = HardwareBufferHandle(hardwareBuffer, /*takeOwnership=*/true),
.size = static_cast<uint32_t>(memory.size),
.name = memory.name,
});
@@ -400,22 +512,7 @@
}
GeneralResult<SharedHandle> unvalidatedConvert(const NativeHandle& aidlNativeHandle) {
- std::vector<base::unique_fd> fds;
- fds.reserve(aidlNativeHandle.fds.size());
- for (const auto& fd : aidlNativeHandle.fds) {
- int dupFd = dup(fd.get());
- if (dupFd == -1) {
- // TODO(b/120417090): is ANEURALNETWORKS_UNEXPECTED_NULL the correct error to return
- // here?
- return NN_ERROR() << "Failed to dup the fd";
- }
- fds.emplace_back(dupFd);
- }
-
- return std::make_shared<const Handle>(Handle{
- .fds = std::move(fds),
- .ints = aidlNativeHandle.ints,
- });
+ return std::make_shared<const Handle>(NN_TRY(unvalidatedConvertHelper(aidlNativeHandle)));
}
GeneralResult<ExecutionPreference> convert(
@@ -508,13 +605,11 @@
return halObject;
}
-} // namespace
-
-nn::GeneralResult<common::NativeHandle> unvalidatedConvert(const nn::SharedHandle& sharedHandle) {
+nn::GeneralResult<common::NativeHandle> unvalidatedConvert(const nn::Handle& handle) {
common::NativeHandle aidlNativeHandle;
- aidlNativeHandle.fds.reserve(sharedHandle->fds.size());
- for (const auto& fd : sharedHandle->fds) {
- int dupFd = dup(fd.get());
+ aidlNativeHandle.fds.reserve(handle.fds.size());
+ for (const auto& fd : handle.fds) {
+ const int dupFd = dup(fd.get());
if (dupFd == -1) {
// TODO(b/120417090): is ANEURALNETWORKS_UNEXPECTED_NULL the correct error to return
// here?
@@ -522,17 +617,69 @@
}
aidlNativeHandle.fds.emplace_back(dupFd);
}
- aidlNativeHandle.ints = sharedHandle->ints;
+ aidlNativeHandle.ints = handle.ints;
return aidlNativeHandle;
}
+static nn::GeneralResult<common::NativeHandle> aidlHandleFromNativeHandle(
+ const native_handle_t& handle) {
+ common::NativeHandle aidlNativeHandle;
+
+ aidlNativeHandle.fds.reserve(handle.numFds);
+ for (int i = 0; i < handle.numFds; ++i) {
+ const int dupFd = dup(handle.data[i]);
+ if (dupFd == -1) {
+ return NN_ERROR(nn::ErrorStatus::GENERAL_FAILURE) << "Failed to dup the fd";
+ }
+ aidlNativeHandle.fds.emplace_back(dupFd);
+ }
+
+ aidlNativeHandle.ints = std::vector<int>(&handle.data[handle.numFds],
+ &handle.data[handle.numFds + handle.numInts]);
+
+ return aidlNativeHandle;
+}
+
+} // namespace
+
+nn::GeneralResult<common::NativeHandle> unvalidatedConvert(const nn::SharedHandle& sharedHandle) {
+ CHECK(sharedHandle != nullptr);
+ return unvalidatedConvert(*sharedHandle);
+}
+
nn::GeneralResult<Memory> unvalidatedConvert(const nn::SharedMemory& memory) {
CHECK(memory != nullptr);
if (memory->size > std::numeric_limits<int64_t>::max()) {
return NN_ERROR() << "Memory size doesn't fit into int64_t.";
}
+ if (const auto* handle = std::get_if<nn::Handle>(&memory->handle)) {
+ return Memory{
+ .handle = NN_TRY(unvalidatedConvert(*handle)),
+ .size = static_cast<int64_t>(memory->size),
+ .name = memory->name,
+ };
+ }
+
+ const auto* ahwb = std::get<nn::HardwareBufferHandle>(memory->handle).get();
+ AHardwareBuffer_Desc bufferDesc;
+ AHardwareBuffer_describe(ahwb, &bufferDesc);
+
+ if (bufferDesc.format == AHARDWAREBUFFER_FORMAT_BLOB) {
+ CHECK_EQ(memory->size, bufferDesc.width);
+ CHECK_EQ(memory->name, "hardware_buffer_blob");
+ } else {
+ CHECK_EQ(memory->size, 0u);
+ CHECK_EQ(memory->name, "hardware_buffer");
+ }
+
+ const native_handle_t* nativeHandle = AHardwareBuffer_getNativeHandle(ahwb);
+ if (nativeHandle == nullptr) {
+ return NN_ERROR() << "unvalidatedConvert failed because AHardwareBuffer_getNativeHandle "
+ "returned nullptr";
+ }
+
return Memory{
- .handle = NN_TRY(unvalidatedConvert(memory->handle)),
+ .handle = NN_TRY(aidlHandleFromNativeHandle(*nativeHandle)),
.size = static_cast<int64_t>(memory->size),
.name = memory->name,
};