gelu vk op tanh fix (#5008)

Tencent · Sep 6, 2023 · e14cc27 · e14cc27
1 parent 4a2620d
commit e14cc27
Show file tree

Hide file tree

Showing 3 changed files with 16 additions and 0 deletions.
diff --git a/src/layer/vulkan/shader/gelu.comp b/src/layer/vulkan/shader/gelu.comp
@@ -62,7 +62,12 @@ void main()
 #endif
 
     // y = 0.5x * (1 + tanh(sqrt(2/Pi) * (x + 0.044715x^3)))
+
+#if NCNN_moltenvk
+    v = 0.5f * v * (1.0f + afp(tanh(float(0.79788452f * (v + 0.044715f * v * v * v)))));
+#else
     v = 0.5f * v * (1.0f + tanh(0.79788452f * (v + 0.044715f * v * v * v)));
+#endif
 
 #if NCNN_image_shader
     image3d_st1(top_blob_3d, ivec3(gx, gy, gz), v);

diff --git a/src/layer/vulkan/shader/gelu_pack4.comp b/src/layer/vulkan/shader/gelu_pack4.comp
@@ -62,7 +62,12 @@ void main()
 #endif
 
     // y = 0.5x * (1 + tanh(sqrt(2/Pi) * (x + 0.044715x^3)))
+
+#if NCNN_moltenvk
+    v = 0.5f * v * (1.0f + afpvec4(tanh(vec4(0.79788452f * (v + 0.044715f * v * v * v)))));
+#else
     v = 0.5f * v * (1.0f + tanh(0.79788452f * (v + 0.044715f * v * v * v)));
+#endif
 
 #if NCNN_image_shader
     image3d_st4(top_blob_3d, ivec3(gx, gy, gz), v);

diff --git a/src/layer/vulkan/shader/gelu_pack8.comp b/src/layer/vulkan/shader/gelu_pack8.comp
@@ -63,8 +63,14 @@ void main()
 #endif
 
     // y = 0.5x * (1 + tanh(sqrt(2/Pi) * (x + 0.044715x^3)))
+
+#if NCNN_moltenvk
+    v[0] = 0.5f * v[0] * (1.0f + afpvec4(tanh(vec4(0.79788452f * (v[0] + 0.044715f * v[0] * v[0] * v[0])))));
+    v[1] = 0.5f * v[1] * (1.0f + afpvec4(tanh(vec4(0.79788452f * (v[1] + 0.044715f * v[1] * v[1] * v[1])))));
+#else
     v[0] = 0.5f * v[0] * (1.0f + tanh(0.79788452f * (v[0] + 0.044715f * v[0] * v[0] * v[0])));
     v[1] = 0.5f * v[1] * (1.0f + tanh(0.79788452f * (v[1] + 0.044715f * v[1] * v[1] * v[1])));
+#endif
 
 #if NCNN_image_shader
     image3d_st8(top_blob_3d, ivec3(gx, gy, gz), v);