ggml : generalize `quantize_fns` for simpler FP16 handling (#1237)

* Generalize quantize_fns for simpler FP16 handling

* Remove call to ggml_cuda_mul_mat_get_wsize

* ci : disable FMA for mac os actions

---------

Co-authored-by: Georgi Gerganov <ggerganov@gmail.com>

1 files changed, 3 insertions, 3 deletions

diff --git a/pocs/vdot/q8dot.cpp b/pocs/vdot/q8dot.cpp
index 5748c8a..4e0e023 100644
--- a/pocs/vdot/q8dot.cpp
+++ b/pocs/vdot/q8dot.cpp
@@ -136,7 +136,7 @@ int main(int argc, char** argv) {
 
     auto ggml_type = type == 0 ? GGML_TYPE_Q4_0 : GGML_TYPE_Q4_1;
 
-    auto funcs = ggml_internal_get_quantize_fn(ggml_type);
+    auto funcs = ggml_internal_get_type_traits(ggml_type);
 
     Stat simple, ggml;
 
@@ -156,8 +156,8 @@ int main(int argc, char** argv) {
 
         t1 = std::chrono::high_resolution_clock::now();
         float fs;
-        if (type == 0) funcs.vec_dot_q(kVecSize * QK4_1, &fs, x40.data(), y.data());
-        else funcs.vec_dot_q(kVecSize * QK4_1, &fs, x41.data(), y.data());
+        if (type == 0) funcs.vec_dot(kVecSize * QK4_1, &fs, x40.data(), y.data());
+        else funcs.vec_dot(kVecSize * QK4_1, &fs, x41.data(), y.data());
         t2 = std::chrono::high_resolution_clock::now();
         t = 1e-3*std::chrono::duration_cast<std::chrono::nanoseconds>(t2-t1).count();
         if (iloop > 3) ggml.addResult(fs, t);