Browse Source

Gelu afp fix (#5039)

tags/20231027
FhqTreap GitHub 2 years ago
parent
commit
a12a14f3a6
No known key found for this signature in database GPG Key ID: 4AEE18F83AFDEB23
3 changed files with 8 additions and 8 deletions
  1. +2
    -2
      src/layer/vulkan/shader/gelu.comp
  2. +2
    -2
      src/layer/vulkan/shader/gelu_pack4.comp
  3. +4
    -4
      src/layer/vulkan/shader/gelu_pack8.comp

+ 2
- 2
src/layer/vulkan/shader/gelu.comp View File

@@ -64,9 +64,9 @@ void main()
// 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)))));
v = afp(0.5f) * v * (afp(1.0f) + afp(tanh(float(afp(0.79788452f) * (v + afp(0.044715f) * v * v * v)))));
#else
v = 0.5f * v * (1.0f + tanh(0.79788452f * (v + 0.044715f * v * v * v)));
v = afp(0.5f) * v * (afp(1.0f) + tanh(afp(0.79788452f) * (v + afp(0.044715f) * v * v * v)));
#endif

#if NCNN_image_shader


+ 2
- 2
src/layer/vulkan/shader/gelu_pack4.comp View File

@@ -64,9 +64,9 @@ void main()
// 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)))));
v = afpvec4(0.5f) * v * (afpvec4(1.0f) + afpvec4(tanh(vec4(afpvec4(0.79788452f) * (v + afpvec4(0.044715f) * v * v * v)))));
#else
v = 0.5f * v * (1.0f + tanh(0.79788452f * (v + 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


+ 4
- 4
src/layer/vulkan/shader/gelu_pack8.comp View File

@@ -65,11 +65,11 @@ void main()
// 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])))));
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] = 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])));
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


Loading…
Cancel
Save