blob: 75dd6dd65647f8e6fc0c5e0f55482f40de8860d6 [file] [log] [blame]
#version 310 es
#extension GL_AMD_gpu_shader_half_float : require
struct Inner {
float scalar_f32;
int scalar_i32;
uint scalar_u32;
float16_t scalar_f16;
vec2 vec2_f32;
ivec2 vec2_i32;
uvec2 vec2_u32;
f16vec2 vec2_f16;
uint pad;
vec3 vec3_f32;
uint pad_1;
ivec3 vec3_i32;
uint pad_2;
uvec3 vec3_u32;
uint pad_3;
f16vec3 vec3_f16;
uint pad_4;
uint pad_5;
vec4 vec4_f32;
ivec4 vec4_i32;
uvec4 vec4_u32;
f16vec4 vec4_f16;
mat2 mat2x2_f32;
uint pad_6;
uint pad_7;
mat2x3 mat2x3_f32;
mat2x4 mat2x4_f32;
mat3x2 mat3x2_f32;
uint pad_8;
uint pad_9;
mat3 mat3x3_f32;
mat3x4 mat3x4_f32;
mat4x2 mat4x2_f32;
mat4x3 mat4x3_f32;
mat4 mat4x4_f32;
f16mat2 mat2x2_f16;
f16mat2x3 mat2x3_f16;
f16mat2x4 mat2x4_f16;
f16mat3x2 mat3x2_f16;
uint pad_10;
f16mat3 mat3x3_f16;
f16mat3x4 mat3x4_f16;
f16mat4x2 mat4x2_f16;
f16mat4x3 mat4x3_f16;
f16mat4 mat4x4_f16;
uint pad_11;
uint pad_12;
vec3 arr2_vec3_f32[2];
f16mat4x2 arr2_mat4x2_f16[2];
};
layout(binding = 0, std430) buffer S_ssbo {
Inner arr[];
} sb;
void assign_and_preserve_padding_sb_arr_X_mat2x3_f32(uint dest[1], mat2x3 value) {
sb.arr[dest[0]].mat2x3_f32[0] = value[0u];
sb.arr[dest[0]].mat2x3_f32[1] = value[1u];
}
void assign_and_preserve_padding_1_sb_arr_X_mat3x3_f32(uint dest[1], mat3 value) {
sb.arr[dest[0]].mat3x3_f32[0] = value[0u];
sb.arr[dest[0]].mat3x3_f32[1] = value[1u];
sb.arr[dest[0]].mat3x3_f32[2] = value[2u];
}
void assign_and_preserve_padding_2_sb_arr_X_mat4x3_f32(uint dest[1], mat4x3 value) {
sb.arr[dest[0]].mat4x3_f32[0] = value[0u];
sb.arr[dest[0]].mat4x3_f32[1] = value[1u];
sb.arr[dest[0]].mat4x3_f32[2] = value[2u];
sb.arr[dest[0]].mat4x3_f32[3] = value[3u];
}
void assign_and_preserve_padding_3_sb_arr_X_mat2x3_f16(uint dest[1], f16mat2x3 value) {
sb.arr[dest[0]].mat2x3_f16[0] = value[0u];
sb.arr[dest[0]].mat2x3_f16[1] = value[1u];
}
void assign_and_preserve_padding_4_sb_arr_X_mat3x3_f16(uint dest[1], f16mat3 value) {
sb.arr[dest[0]].mat3x3_f16[0] = value[0u];
sb.arr[dest[0]].mat3x3_f16[1] = value[1u];
sb.arr[dest[0]].mat3x3_f16[2] = value[2u];
}
void assign_and_preserve_padding_5_sb_arr_X_mat4x3_f16(uint dest[1], f16mat4x3 value) {
sb.arr[dest[0]].mat4x3_f16[0] = value[0u];
sb.arr[dest[0]].mat4x3_f16[1] = value[1u];
sb.arr[dest[0]].mat4x3_f16[2] = value[2u];
sb.arr[dest[0]].mat4x3_f16[3] = value[3u];
}
void assign_and_preserve_padding_6_sb_arr_X_arr2_vec3_f32(uint dest[1], vec3 value[2]) {
{
for(uint i = 0u; (i < 2u); i = (i + 1u)) {
sb.arr[dest[0]].arr2_vec3_f32[i] = value[i];
}
}
}
void tint_symbol(uint idx) {
sb.arr[idx].scalar_f32 = 0.0f;
sb.arr[idx].scalar_i32 = 0;
sb.arr[idx].scalar_u32 = 0u;
sb.arr[idx].scalar_f16 = 0.0hf;
sb.arr[idx].vec2_f32 = vec2(0.0f);
sb.arr[idx].vec2_i32 = ivec2(0);
sb.arr[idx].vec2_u32 = uvec2(0u);
sb.arr[idx].vec2_f16 = f16vec2(0.0hf);
sb.arr[idx].vec3_f32 = vec3(0.0f);
sb.arr[idx].vec3_i32 = ivec3(0);
sb.arr[idx].vec3_u32 = uvec3(0u);
sb.arr[idx].vec3_f16 = f16vec3(0.0hf);
sb.arr[idx].vec4_f32 = vec4(0.0f);
sb.arr[idx].vec4_i32 = ivec4(0);
sb.arr[idx].vec4_u32 = uvec4(0u);
sb.arr[idx].vec4_f16 = f16vec4(0.0hf);
sb.arr[idx].mat2x2_f32 = mat2(vec2(0.0f), vec2(0.0f));
uint tint_symbol_1[1] = uint[1](idx);
assign_and_preserve_padding_sb_arr_X_mat2x3_f32(tint_symbol_1, mat2x3(vec3(0.0f), vec3(0.0f)));
sb.arr[idx].mat2x4_f32 = mat2x4(vec4(0.0f), vec4(0.0f));
sb.arr[idx].mat3x2_f32 = mat3x2(vec2(0.0f), vec2(0.0f), vec2(0.0f));
uint tint_symbol_2[1] = uint[1](idx);
assign_and_preserve_padding_1_sb_arr_X_mat3x3_f32(tint_symbol_2, mat3(vec3(0.0f), vec3(0.0f), vec3(0.0f)));
sb.arr[idx].mat3x4_f32 = mat3x4(vec4(0.0f), vec4(0.0f), vec4(0.0f));
sb.arr[idx].mat4x2_f32 = mat4x2(vec2(0.0f), vec2(0.0f), vec2(0.0f), vec2(0.0f));
uint tint_symbol_3[1] = uint[1](idx);
assign_and_preserve_padding_2_sb_arr_X_mat4x3_f32(tint_symbol_3, mat4x3(vec3(0.0f), vec3(0.0f), vec3(0.0f), vec3(0.0f)));
sb.arr[idx].mat4x4_f32 = mat4(vec4(0.0f), vec4(0.0f), vec4(0.0f), vec4(0.0f));
sb.arr[idx].mat2x2_f16 = f16mat2(f16vec2(0.0hf), f16vec2(0.0hf));
uint tint_symbol_4[1] = uint[1](idx);
assign_and_preserve_padding_3_sb_arr_X_mat2x3_f16(tint_symbol_4, f16mat2x3(f16vec3(0.0hf), f16vec3(0.0hf)));
sb.arr[idx].mat2x4_f16 = f16mat2x4(f16vec4(0.0hf), f16vec4(0.0hf));
sb.arr[idx].mat3x2_f16 = f16mat3x2(f16vec2(0.0hf), f16vec2(0.0hf), f16vec2(0.0hf));
uint tint_symbol_5[1] = uint[1](idx);
assign_and_preserve_padding_4_sb_arr_X_mat3x3_f16(tint_symbol_5, f16mat3(f16vec3(0.0hf), f16vec3(0.0hf), f16vec3(0.0hf)));
sb.arr[idx].mat3x4_f16 = f16mat3x4(f16vec4(0.0hf), f16vec4(0.0hf), f16vec4(0.0hf));
sb.arr[idx].mat4x2_f16 = f16mat4x2(f16vec2(0.0hf), f16vec2(0.0hf), f16vec2(0.0hf), f16vec2(0.0hf));
uint tint_symbol_6[1] = uint[1](idx);
assign_and_preserve_padding_5_sb_arr_X_mat4x3_f16(tint_symbol_6, f16mat4x3(f16vec3(0.0hf), f16vec3(0.0hf), f16vec3(0.0hf), f16vec3(0.0hf)));
sb.arr[idx].mat4x4_f16 = f16mat4(f16vec4(0.0hf), f16vec4(0.0hf), f16vec4(0.0hf), f16vec4(0.0hf));
uint tint_symbol_7[1] = uint[1](idx);
vec3 tint_symbol_8[2] = vec3[2](vec3(0.0f), vec3(0.0f));
assign_and_preserve_padding_6_sb_arr_X_arr2_vec3_f32(tint_symbol_7, tint_symbol_8);
f16mat4x2 tint_symbol_9[2] = f16mat4x2[2](f16mat4x2(f16vec2(0.0hf), f16vec2(0.0hf), f16vec2(0.0hf), f16vec2(0.0hf)), f16mat4x2(f16vec2(0.0hf), f16vec2(0.0hf), f16vec2(0.0hf), f16vec2(0.0hf)));
sb.arr[idx].arr2_mat4x2_f16 = tint_symbol_9;
}
layout(local_size_x = 1, local_size_y = 1, local_size_z = 1) in;
void main() {
tint_symbol(gl_LocalInvocationIndex);
return;
}