happyz
/
gemma.cpp
mirror of https://github.com/google/gemma.cpp.git
1
0
Fork 0
Code Issues Packages Projects Releases Wiki Activity

revert back to HWY_ASSERT for lane constraints, qualify hn::Add 

PiperOrigin-RevId: 640193239
This commit is contained in:
Paul Chang 2024-06-04 10:09:26 -07:00 committed by Copybara-Service
parent e71d82ead9
commit 175e389c3c
2 changed files with 3 additions and 3 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

2
compression/nuq-inl.h
View File

@ -207,7 +207,7 @@ class NuqClustering {
VI arg = hn::Zero(di);
// For each j (start of rightmost cluster):
VI vj = k1;
for (size_t j = 1; j < last + N; ++j, vj = Add(vj, k1)) {
for (size_t j = 1; j < last + N; ++j, vj = hn::Add(vj, k1)) {
const VF c = ClusterDynProg(df, D, cc, num_clusters, last, j);
// Retain the min cost and the j index that caused it.

4
gemma/ops.h
View File

@ -371,8 +371,8 @@ void GEMM_4x4_Static(const MatT* HWY_RESTRICT A, const MatT* HWY_RESTRICT B,
constexpr size_t kRegCols = 4; // in vectors
static_assert(kRowsAC % kRegRows == 0);
static_assert(kColsA_RowsB % (N * kRegCols) == 0);
static_assert(kColsBC % kRegCols == 0);
HWY_ASSERT(kColsA_RowsB % (N * kRegCols) == 0);
constexpr size_t kTilesY = kRowsAC / kRegRows;
constexpr size_t kTilesX = kColsBC / kRegCols;
constexpr size_t kTiles = kTilesX * kTilesY;
@ -405,8 +405,8 @@ HWY_NOINLINE void MatMul_4x4(const MatT* HWY_RESTRICT A,
constexpr size_t kRegCols = 4; // in vectors
static_assert(kRowsAC % kRegRows == 0);
static_assert(kColsA_RowsB % (N * kRegCols) == 0);
static_assert(kColsBC % kRegCols == 0);
HWY_ASSERT(kColsA_RowsB % (N * kRegCols) == 0);
const size_t kTilesY = kRowsAC / kRegRows;
const size_t kTilesX = kColsBC / kRegCols;
const size_t kTiles = kTilesX * kTilesY;