more refinement to use fma for mul_add

8a7e325c · Shucai Xiao · fdc0ae82 · 8a7e325c
Commit 8a7e325c authored Mar 03, 2022 by Shucai Xiao
Hide whitespace changes
Inline Side-by-side

Showing with 1 addition and 1 deletion

src/targets/gpu/device/mul_add.cpp src/targets/gpu/device/mul_add.cpp +1 -1

No files found.
--- a/src/targets/gpu/device/mul_add.cpp
+++ b/src/targets/gpu/device/mul_add.cpp
@@ -21,7 +21,7 @@ __global__ void mul_add_kernel_dim3(void* a, void* x, void* b, int dim3, void* r
    if(id < n)
    {
        auto id1 = id % dim3;
-        hr[id]   = __hadd2(__hmul2(ha[id], hx[id1]), hb[id1]);
+        hr[id]   = __hfma2(ha[id], hx[id1], hb[id1]);
    }
 }