Commit fdc0ae82 authored by Shucai Xiao's avatar Shucai Xiao
Browse files

clang format

parent 6c834296
...@@ -59,12 +59,12 @@ __global__ void add_gelu_kernel(void* a, void* b, int n_dim, void* r, int n) ...@@ -59,12 +59,12 @@ __global__ void add_gelu_kernel(void* a, void* b, int n_dim, void* r, int n)
__half2 sqrt2 = __float2half2_rn(M_SQRT1_2); __half2 sqrt2 = __float2half2_rn(M_SQRT1_2);
auto x = __hmul2(sum, sqrt2); auto x = __hmul2(sum, sqrt2);
auto f2 = __half22float2(x); auto f2 = __half22float2(x);
f2.x = ::erf(f2.x); f2.x = ::erf(f2.x);
f2.y = ::erf(f2.y); f2.y = ::erf(f2.y);
auto h2 = __floats2half2_rn(f2.x, f2.y); auto h2 = __floats2half2_rn(f2.x, f2.y);
auto one = __float2half2_rn(1.0f); auto one = __float2half2_rn(1.0f);
h2 = __hadd2(h2, one); h2 = __hadd2(h2, one);
__half2 point5 = __float2half2_rn(0.5f); __half2 point5 = __float2half2_rn(0.5f);
hr[tid] = __hmul2(sum, __hmul2(point5, h2)); hr[tid] = __hmul2(sum, __hmul2(point5, h2));
......
Markdown is supported
0% or .
You are about to add 0 people to the discussion. Proceed with caution.
Finish editing this message first!
Please register or to comment