intel/compute-runtime
1
0
Fork 0
mirror of https://github.com/intel/compute-runtime.git synced 2026-01-07 21:27:04 +08:00
Code Issues Packages Projects Releases Wiki Activity

[perf] simplify setting constant buffer and improve performance

Browse Source 
- no need to count parameters
- remove unrecoverable which requires fetching additional fields.

Related-To: NEO-5170

Signed-off-by: Michal Mrozek <michal.mrozek@intel.com>
This commit is contained in:
Michal Mrozek
2023-03-02 11:09:24 +00:00
committed by Compute-Runtime-Automation
parent 49424eb859
commit c77d954900
2 changed files with 7 additions and 16 deletions
Download Patch File Download Diff File Expand all files Collapse all files

11
shared/source/kernel/kernel_arg_descriptor.h
View File

@@ -1,5 +1,5 @@
/*
* Copyright (C) 2020-2022 Intel Corporation
* Copyright (C) 2020-2023 Intel Corporation
*
* SPDX-License-Identifier: MIT
*
@@ -282,18 +282,17 @@ inline bool patchNonPointer(ArrayRef<uint8_t> buffer, CrossThreadDataOffset loca
if (undefined<CrossThreadDataOffset> == location) {
return false;
}
UNRECOVERABLE_IF(location + sizeof(DstT) > buffer.size());
DEBUG_BREAK_IF(location + sizeof(DstT) > buffer.size());
*reinterpret_cast<DstT *>(buffer.begin() + location) = static_cast<DstT>(value);
return true;
}
template <uint32_t VecSize, typename T>
inline uint32_t patchVecNonPointer(ArrayRef<uint8_t> buffer, const CrossThreadDataOffset (&location)[VecSize], const T (&value)[VecSize]) {
uint32_t numPatched = 0;
inline void patchVecNonPointer(ArrayRef<uint8_t> buffer, const CrossThreadDataOffset (&location)[VecSize], const T (&value)[VecSize]) {
for (uint32_t i = 0; i < VecSize; ++i) {
numPatched += patchNonPointer<T, T>(buffer, location[i], value[i]) ? 1 : 0;
patchNonPointer<T, T>(buffer, location[i], value[i]);
}
return numPatched;
return;
}
inline bool patchPointer(ArrayRef<uint8_t> buffer, const ArgDescPointer &arg, uintptr_t value) {