Update progress

AyiStar · Jul 27, 2024 · 04a2ca4 · 04a2ca4
1 parent e92ad41
commit 04a2ca4
Show file tree

Hide file tree

Showing 3 changed files with 11 additions and 9 deletions.
diff --git a/dev.md b/dev.md
@@ -97,7 +97,7 @@ Output a well-organized technical report.
 #### Benchmarks
 - [x] Add arguments with different types.
 - [x] Prepare more model weights.
-- [ ] Benchmark more models.
+- [x] Benchmark more models.
 
 #### Autotune
 TBD
diff --git a/src/lamm_kernel_q2_k.hpp b/src/lamm_kernel_q2_k.hpp
@@ -167,7 +167,8 @@ LA_INLINE void lamm_simd_block_kernel(const block_q2_K *a, const block_q8_K *b,
 
   static_assert(B0 > 0 && B0 <= 4);
   static_assert(B1 > 0 && B1 <= 4);
-  // std::cout << "Q2_K SIMD block called with B0=" << B0 << ", B1=" << B1 << std::endl;
+  // std::cout << "Q2_K SIMD block called with B0=" << B0 << ", B1=" << B1 <<
+  // std::endl;
   using namespace simd;
 
   const ivreg_t m3 = ivset(3);
@@ -281,12 +282,12 @@ LA_INLINE void lamm_simd_block_kernel(const block_q2_K *a, const block_q8_K *b,
 #undef INNER_FN
 #undef OUTER_FN
 
-#define FN(N)                                                                 \
-  if constexpr (B0 > N) {                                                     \
-    ai##N++; \
-  } \
-  if constexpr (B1 > N) {                                                     \
-    bj##N++; \
+#define FN(N)                                                                  \
+  if constexpr (B0 > N) {                                                      \
+    ai##N++;                                                                   \
+  }                                                                            \
+  if constexpr (B1 > N) {                                                      \
+    bj##N++;                                                                   \
   }
     LOOP(FN, 4)
 #undef FN

diff --git a/src/loongarch_matmul.cpp b/src/loongarch_matmul.cpp
@@ -98,7 +98,8 @@ void lamm_mul_mat(const struct ggml_compute_params *params,
   C.col = ne11;
   C.ld = nb1 / ggml_type_size(dst->type);
 
-  using MatMulFuncPtr = void(*)(const Matrix &A, const Matrix &B, const Matrix &C, int ith, int nth);
+  using MatMulFuncPtr = void (*)(const Matrix &A, const Matrix &B,
+                                 const Matrix &C, int ith, int nth);
   MatMulFuncPtr mm_func = nullptr;
   switch (A.type) {
   case GGML_TYPE_F32: