afpvec4 fix

Tencent · Sep 18, 2023 · af70ca9 · af70ca9
1 parent b9a9d85
commit af70ca9
Show file tree

Hide file tree

Showing 2 changed files with 6 additions and 6 deletions.
diff --git a/src/layer/vulkan/shader/gelu_pack4.comp b/src/layer/vulkan/shader/gelu_pack4.comp
@@ -64,9 +64,9 @@ void main()
     // y = 0.5x * (1 + tanh(sqrt(2/Pi) * (x + 0.044715x^3)))
 
 #if NCNN_moltenvk
-    v = afp(0.5f) * v * (afp(1.0f) + afpvec4(tanh(vec4(afp(0.79788452f) * (v + afp(0.044715f) * v * v * v)))));
+    v = afpvec4(0.5f) * v * (afpvec4(1.0f) + afpvec4(tanh(vec4(afpvec4(0.79788452f) * (v + afpvec4(0.044715f) * v * v * v)))));
 #else
-    v = afp(0.5f) * v * (afp(1.0f) + tanh(afp(0.79788452f) * (v + afp(0.044715f) * v * v * v)));
+    v = afpvec4(0.5f) * v * (afpvec4(1.0f) + tanh(afpvec4(0.79788452f) * (v + afpvec4(0.044715f) * v * v * v)));
 #endif
 
 #if NCNN_image_shader

diff --git a/src/layer/vulkan/shader/gelu_pack8.comp b/src/layer/vulkan/shader/gelu_pack8.comp
@@ -65,11 +65,11 @@ void main()
     // y = 0.5x * (1 + tanh(sqrt(2/Pi) * (x + 0.044715x^3)))
 
 #if NCNN_moltenvk
-    v[0] = afp(0.5f) * v[0] * (afp(1.0f) + afpvec4(tanh(vec4(afp(0.79788452f) * (v[0] + afp(0.044715f) * v[0] * v[0] * v[0])))));
-    v[1] = afp(0.5f) * v[1] * (afp(1.0f) + afpvec4(tanh(vec4(afp(0.79788452f) * (v[1] + afp(0.044715f) * v[1] * v[1] * v[1])))));
+    v[0] = afpvec4(0.5f) * v[0] * (afpvec4(1.0f) + afpvec4(tanh(vec4(afpvec4(0.79788452f) * (v[0] + afpvec4(0.044715f) * v[0] * v[0] * v[0])))));
+    v[1] = afpvec4(0.5f) * v[1] * (afpvec4(1.0f) + afpvec4(tanh(vec4(afpvec4(0.79788452f) * (v[1] + afpvec4(0.044715f) * v[1] * v[1] * v[1])))));
 #else
-    v[0] = afp(0.5f) * v[0] * (afp(1.0f) + tanh(afp(0.79788452f) * (v[0] + afp(0.044715f) * v[0] * v[0] * v[0])));
-    v[1] = afp(0.5f) * v[1] * (afp(1.0f) + tanh(afp(0.79788452f) * (v[1] + afp(0.044715f) * v[1] * v[1] * v[1])));
+    v[0] = afpvec4(0.5f) * v[0] * (afpvec4(1.0f) + tanh(afpvec4(0.79788452f) * (v[0] + afpvec4(0.044715f) * v[0] * v[0] * v[0])));
+    v[1] = afpvec4(0.5f) * v[1] * (afpvec4(1.0f) + tanh(afpvec4(0.79788452f) * (v[1] + afpvec4(0.044715f) * v[1] * v[1] * v[1])));
 #endif
 
 #if NCNN_image_shader