thek0tyara/llama-cpp-turboquant
1
0
Fork 0
Code Issues Pull requests Projects Releases Packages Wiki Activity Actions 4

Reformat

Browse source
This commit is contained in:
Alejandro Saucedo 2021-03-13 17:02:23 +00:00
parent 53ae5c2d0a
commit ea3a303984
5 changed files with 89 additions and 88 deletions
Download patch file Download diff file Expand all files Collapse all files

47
src/OpMemoryBarrier.cpp
View file

@ -5,21 +5,20 @@
namespace kp {
OpMemoryBarrier::OpMemoryBarrier(
const std::vector<std::shared_ptr<Tensor>>& tensors,
const vk::AccessFlagBits& srcAccessMask,
const vk::AccessFlagBits& dstAccessMask,
const vk::PipelineStageFlagBits& srcStageMask,
const vk::PipelineStageFlagBits& dstStageMask,
bool barrierOnPrimary)
: mTensors(tensors),
mSrcAccessMask(srcAccessMask),
mDstAccessMask(dstAccessMask),
mSrcStageMask(srcStageMask),
mDstStageMask(dstStageMask),
mBarrierOnPrimary(barrierOnPrimary)
const std::vector<std::shared_ptr<Tensor>>& tensors,
const vk::AccessFlagBits& srcAccessMask,
const vk::AccessFlagBits& dstAccessMask,
const vk::PipelineStageFlagBits& srcStageMask,
const vk::PipelineStageFlagBits& dstStageMask,
bool barrierOnPrimary)
: mTensors(tensors)
, mSrcAccessMask(srcAccessMask)
, mDstAccessMask(dstAccessMask)
, mSrcStageMask(srcStageMask)
, mDstStageMask(dstStageMask)
, mBarrierOnPrimary(barrierOnPrimary)
{
KP_LOG_DEBUG("Kompute OpMemoryBarrier constructor");
}
OpMemoryBarrier::~OpMemoryBarrier()
@ -35,21 +34,19 @@ OpMemoryBarrier::record(const vk::CommandBuffer& commandBuffer)
// Barrier to ensure the data is finished writing to buffer memory
if (this->mBarrierOnPrimary) {
for (const std::shared_ptr<Tensor>& tensor : this->mTensors) {
tensor->recordPrimaryBufferMemoryBarrier(
commandBuffer,
this->mSrcAccessMask,
this->mDstAccessMask,
this->mSrcStageMask,
this->mDstStageMask);
tensor->recordPrimaryBufferMemoryBarrier(commandBuffer,
this->mSrcAccessMask,
this->mDstAccessMask,
this->mSrcStageMask,
this->mDstStageMask);
}
} else {
for (const std::shared_ptr<Tensor>& tensor : this->mTensors) {
tensor->recordStagingBufferMemoryBarrier(
commandBuffer,
this->mSrcAccessMask,
this->mDstAccessMask,
this->mSrcStageMask,
this->mDstStageMask);
tensor->recordStagingBufferMemoryBarrier(commandBuffer,
this->mSrcAccessMask,
this->mDstAccessMask,
this->mSrcStageMask,
this->mDstStageMask);
}
}
}

3
src/OpTensorCopy.cpp
View file

@ -44,8 +44,7 @@ OpTensorCopy::record(const vk::CommandBuffer& commandBuffer)
// We iterate from the second tensor onwards and record a copy to all
for (size_t i = 1; i < this->mTensors.size(); i++) {
this->mTensors[i]->recordCopyFrom(
commandBuffer, this->mTensors[0]);
this->mTensors[i]->recordCopyFrom(commandBuffer, this->mTensors[0]);
}
}

22
src/OpTensorSyncLocal.cpp
View file

@ -31,19 +31,21 @@ OpTensorSyncLocal::record(const vk::CommandBuffer& commandBuffer)
for (size_t i = 0; i < this->mTensors.size(); i++) {
if (this->mTensors[i]->tensorType() == Tensor::TensorTypes::eDevice) {
this->mTensors[i]->recordPrimaryBufferMemoryBarrier(commandBuffer,
vk::AccessFlagBits::eShaderWrite,
vk::AccessFlagBits::eTransferRead,
vk::PipelineStageFlagBits::eComputeShader,
vk::PipelineStageFlagBits::eTransfer);
this->mTensors[i]->recordPrimaryBufferMemoryBarrier(
commandBuffer,
vk::AccessFlagBits::eShaderWrite,
vk::AccessFlagBits::eTransferRead,
vk::PipelineStageFlagBits::eComputeShader,
vk::PipelineStageFlagBits::eTransfer);
this->mTensors[i]->recordCopyFromDeviceToStaging(commandBuffer);
this->mTensors[i]->recordPrimaryBufferMemoryBarrier(commandBuffer,
vk::AccessFlagBits::eTransferWrite,
vk::AccessFlagBits::eHostRead,
vk::PipelineStageFlagBits::eTransfer,
vk::PipelineStageFlagBits::eHost);
this->mTensors[i]->recordPrimaryBufferMemoryBarrier(
commandBuffer,
vk::AccessFlagBits::eTransferWrite,
vk::AccessFlagBits::eHostRead,
vk::PipelineStageFlagBits::eTransfer,
vk::PipelineStageFlagBits::eHost);
}
}
}

39
src/Tensor.cpp
View file

@ -131,8 +131,7 @@ Tensor::mapRawData()
this->mRawData = this->mDevice->mapMemory(
*hostVisibleMemory, 0, bufferSize, vk::MemoryMapFlags());
vk::MappedMemoryRange mappedMemoryRange(
*hostVisibleMemory, 0, bufferSize);
vk::MappedMemoryRange mappedMemoryRange(*hostVisibleMemory, 0, bufferSize);
}
void
@ -219,36 +218,36 @@ Tensor::recordCopyBuffer(const vk::CommandBuffer& commandBuffer,
void
Tensor::recordPrimaryBufferMemoryBarrier(const vk::CommandBuffer& commandBuffer,
vk::AccessFlagBits srcAccessMask,
vk::AccessFlagBits dstAccessMask,
vk::PipelineStageFlagBits srcStageMask,
vk::PipelineStageFlagBits dstStageMask)
vk::AccessFlagBits srcAccessMask,
vk::AccessFlagBits dstAccessMask,
vk::PipelineStageFlagBits srcStageMask,
vk::PipelineStageFlagBits dstStageMask)
{
KP_LOG_DEBUG("Kompute Tensor recording PRIMARY buffer memory barrier");
this->recordBufferMemoryBarrier(commandBuffer,
*this->mPrimaryBuffer,
srcAccessMask,
dstAccessMask,
srcStageMask,
dstStageMask);
*this->mPrimaryBuffer,
srcAccessMask,
dstAccessMask,
srcStageMask,
dstStageMask);
}
void
Tensor::recordStagingBufferMemoryBarrier(const vk::CommandBuffer& commandBuffer,
vk::AccessFlagBits srcAccessMask,
vk::AccessFlagBits dstAccessMask,
vk::PipelineStageFlagBits srcStageMask,
vk::PipelineStageFlagBits dstStageMask)
vk::AccessFlagBits srcAccessMask,
vk::AccessFlagBits dstAccessMask,
vk::PipelineStageFlagBits srcStageMask,
vk::PipelineStageFlagBits dstStageMask)
{
KP_LOG_DEBUG("Kompute Tensor recording PRIMARY buffer memory barrier");
this->recordBufferMemoryBarrier(commandBuffer,
*this->mStagingBuffer,
srcAccessMask,
dstAccessMask,
srcStageMask,
dstStageMask);
*this->mStagingBuffer,
srcAccessMask,
dstAccessMask,
srcStageMask,
dstStageMask);
}
void

66
src/include/kompute/Tensor.hpp
View file

@ -120,8 +120,9 @@ class Tensor
void recordCopyFromDeviceToStaging(const vk::CommandBuffer& commandBuffer);
/**
* Records the buffer memory barrier into the primary buffer and command buffer which
* ensures that relevant data transfers are carried out correctly.
* Records the buffer memory barrier into the primary buffer and command
* buffer which ensures that relevant data transfers are carried out
* correctly.
*
* @param commandBuffer Vulkan Command Buffer to record the commands into
* @param srcAccessMask Access flags for source access mask
@ -129,14 +130,16 @@ class Tensor
* @param scrStageMask Pipeline stage flags for source stage mask
* @param dstStageMask Pipeline stage flags for destination stage mask
*/
void recordPrimaryBufferMemoryBarrier(const vk::CommandBuffer& commandBuffer,
vk::AccessFlagBits srcAccessMask,
vk::AccessFlagBits dstAccessMask,
vk::PipelineStageFlagBits srcStageMask,
vk::PipelineStageFlagBits dstStageMask);
void recordPrimaryBufferMemoryBarrier(
const vk::CommandBuffer& commandBuffer,
vk::AccessFlagBits srcAccessMask,
vk::AccessFlagBits dstAccessMask,
vk::PipelineStageFlagBits srcStageMask,
vk::PipelineStageFlagBits dstStageMask);
/**
* Records the buffer memory barrier into the staging buffer and command buffer which
* ensures that relevant data transfers are carried out correctly.
* Records the buffer memory barrier into the staging buffer and command
* buffer which ensures that relevant data transfers are carried out
* correctly.
*
* @param commandBuffer Vulkan Command Buffer to record the commands into
* @param srcAccessMask Access flags for source access mask
@ -144,12 +147,12 @@ class Tensor
* @param scrStageMask Pipeline stage flags for source stage mask
* @param dstStageMask Pipeline stage flags for destination stage mask
*/
void recordStagingBufferMemoryBarrier(const vk::CommandBuffer& commandBuffer,
vk::AccessFlagBits srcAccessMask,
vk::AccessFlagBits dstAccessMask,
vk::PipelineStageFlagBits srcStageMask,
vk::PipelineStageFlagBits dstStageMask);
void recordStagingBufferMemoryBarrier(
const vk::CommandBuffer& commandBuffer,
vk::AccessFlagBits srcAccessMask,
vk::AccessFlagBits dstAccessMask,
vk::PipelineStageFlagBits srcStageMask,
vk::PipelineStageFlagBits dstStageMask);
/**
* Constructs a vulkan descriptor buffer info which can be used to specify
@ -172,17 +175,17 @@ class Tensor
* Returns the total size of a single element of the respective data type
* that this tensor holds.
*
* @return Unsigned integer representing the memory of a single element of the
* respective data type.
* @return Unsigned integer representing the memory of a single element of
* the respective data type.
*/
uint32_t dataTypeMemorySize();
/**
* Returns the total memory size of the data contained by the Tensor object which
* would equate to (this->size() * this->dataTypeMemorySize())
* Returns the total memory size of the data contained by the Tensor object
* which would equate to (this->size() * this->dataTypeMemorySize())
*
* @return Unsigned integer representing the memory of a single element of the
* respective data type.
* @return Unsigned integer representing the memory of a single element of
* the respective data type.
*/
uint32_t memorySize();
@ -194,23 +197,24 @@ class Tensor
TensorDataTypes dataType();
/**
* Retrieve the raw data via the pointer to the memory that contains the raw memory
* of this current tensor. This tensor gets changed to a nullptr when the Tensor is
* removed.
* Retrieve the raw data via the pointer to the memory that contains the raw
* memory of this current tensor. This tensor gets changed to a nullptr when
* the Tensor is removed.
*
* @return Pointer to raw memory containing raw bytes data of Tensor.
*/
void* rawData();
/**
* Sets / resets the data of the tensor which is directly done on the GPU host visible
* memory available by the tensor.
* Sets / resets the data of the tensor which is directly done on the GPU
* host visible memory available by the tensor.
*/
void setRawData(const void* data);
/**
* Template to return the pointer data converted by specific type, which would be
* any of the supported types including float, double, int32, uint32 and bool.
* Template to return the pointer data converted by specific type, which
* would be any of the supported types including float, double, int32,
* uint32 and bool.
*
* @return Pointer to raw memory containing raw bytes data of Tensor.
*/
@ -221,8 +225,9 @@ class Tensor
}
/**
* Template to get the data of the current tensor as a vector of specific type, which would be
* any of the supported types including float, double, int32, uint32 and bool.
* Template to get the data of the current tensor as a vector of specific
* type, which would be any of the supported types including float, double,
* int32, uint32 and bool.
*
* @return Vector of type provided by template.
*/
@ -241,7 +246,6 @@ class Tensor
void* mRawData;
private:
// -------------- NEVER OWNED RESOURCES
std::shared_ptr<vk::PhysicalDevice> mPhysicalDevice;
std::shared_ptr<vk::Device> mDevice;