potok/opencv
1
0
Fork 0
mirror of https://github.com/opencv/opencv.git synced 2026-07-21 19:33:03 +04:00
Code Packages Releases Wiki Activity

Fixed MLAS build with older versions of GCC.

Browse Source
This commit is contained in:
Alexander Smorkalov
2026-05-26 11:23:46 +03:00
parent be2c53c2b7
commit f251467b55
3 changed files with 7 additions and 5 deletions
Download Patch File Download Diff File Expand all files Collapse all files
3rdparty/mlas/lib/mlasi.h
Vendored
+3 -3
View File
@@ -313,7 +313,7 @@ operator!=(const MLFloat16& left, const MLFloat16& right)
#endif // BUILD_MLAS_NO_ONNXRUNTIME
static_assert(sizeof(MLAS_FP16) == FP16_SIZE);
static_assert(sizeof(MLAS_FP16) == FP16_SIZE, "");
//
@@ -3080,7 +3080,7 @@ MLAS_FORCEINLINE
void
MlasSetInt4Element(uint8_t* Output, size_t ElemIndex, UnpackedType Value)
{
static_assert(std::is_same_v<UnpackedType, uint8_t> || std::is_same_v<UnpackedType, int8_t>);
static_assert(std::is_same<UnpackedType, uint8_t>::value || std::is_same<UnpackedType, int8_t>::value, "");
const size_t OutputIndex = ElemIndex >> 1; // which byte
const size_t NibbleIndex = ElemIndex & 0x1; // which 4-bit elem in the byte
@@ -3097,6 +3097,6 @@ MLAS_FORCEINLINE
void
MlasPackInt4Elements(uint8_t* Output, UnpackedType ValueLow, UnpackedType ValueHigh)
{
static_assert(std::is_same_v<UnpackedType, uint8_t> || std::is_same_v<UnpackedType, int8_t>);
static_assert(std::is_same<UnpackedType, uint8_t>::value || std::is_same<UnpackedType, int8_t>::value, "");
*Output = static_cast<uint8_t>(((ValueHigh & 0xF) << 4) | (ValueLow & 0xF));
}
modules/dnn/src/graph_fusion_attention.cpp
+2 -1
View File
@@ -335,7 +335,8 @@ struct ModelFusionAttention
bool modified = false;
std::set<int> removed_ops;
for (auto& [inp_idx, matmul_indices] : qkv_candidates) {
for (auto& candidate : qkv_candidates) {
auto matmul_indices = candidate.second;
if (matmul_indices.size() < 3) continue;
for (size_t qi = 0; qi + 2 < matmul_indices.size(); qi++)
modules/dnn/src/graph_fusion_shared_gemm.cpp
+2 -1
View File
@@ -147,7 +147,8 @@ struct ModelFusionSharedGemm
std::set<int> removed_ops;
vector<std::pair<int, vector<Ptr<Layer>>>> insertions; // (insert_pos, fused-and-slice layers)
for (auto& [key, infos] : groups) {
for (auto& group : groups) {
auto infos = group.second;
if (infos.size() < 2) continue;
bool all_have_bias = infos[0].has_bias;