mirror of
https://github.com/RPCS3/glslang.git
synced 2024-11-28 13:40:41 +00:00
1189a7bc4a
The 310 spec (and desktop specs) have clarified this is a waring, not an error, but 300 tests still expect an error.
77 lines
1.5 KiB
GLSL
77 lines
1.5 KiB
GLSL
#version 300 es
|
|
|
|
int imax, imin;
|
|
uint umax, umin;
|
|
|
|
vec3 x, y; // ERROR, needs default precision
|
|
bvec3 bv;
|
|
|
|
uint uy;
|
|
uvec2 uv2c;
|
|
uvec2 uv2y;
|
|
uvec2 uv2x;
|
|
uvec4 uv4y;
|
|
|
|
ivec3 iv3a;
|
|
ivec3 iv3b;
|
|
|
|
ivec4 iv4a;
|
|
ivec4 iv4b;
|
|
|
|
float f;
|
|
|
|
vec2 v2a, v2b;
|
|
vec4 v4;
|
|
|
|
void main()
|
|
{
|
|
// 1.3 int
|
|
vec3 v = mix(x, y, bv);
|
|
ivec4 iv10 = abs(iv4a);
|
|
ivec4 iv11 = sign(iv4a);
|
|
ivec4 iv12 = min(iv4a, iv4b);
|
|
ivec4 iv13 = min(iv4a, imin);
|
|
uvec2 u = min(uv2x, uv2y);
|
|
uvec4 uv = min(uv4y, uy);
|
|
ivec3 iv14 = max(iv3a, iv3b);
|
|
ivec4 iv15 = max(iv4a, imax);
|
|
uvec2 u10 = max(uv2x, uv2y);
|
|
uvec2 u11 = max(uv2x, uy);
|
|
ivec4 iv16 = clamp(iv4a, iv4a, iv4b);
|
|
ivec4 iv17 = clamp(iv4a, imin, imax);
|
|
uvec2 u12 = clamp(uv2x, uv2y, uv2c);
|
|
uvec4 uv10 = clamp(uv4y, umin, umax);
|
|
|
|
// 1.3 float
|
|
vec3 modfOut;
|
|
vec3 v11 = modf(x, modfOut);
|
|
|
|
float t = trunc(f);
|
|
vec2 v12 = round(v2a);
|
|
vec2 v13 = roundEven(v2a);
|
|
bvec2 b10 = isnan(v2a);
|
|
bvec4 b11 = isinf(v4);
|
|
|
|
// 3.3 float
|
|
int i = floatBitsToInt(f);
|
|
uvec4 uv11 = floatBitsToUint(v4);
|
|
vec4 v14 = intBitsToFloat(iv4a);
|
|
vec2 v15 = uintBitsToFloat(uv2c);
|
|
|
|
// 4.0 pack
|
|
uint u19 = packSnorm2x16(v2a);
|
|
vec2 v20 = unpackSnorm2x16(uy);
|
|
uint u15 = packUnorm2x16(v2a);
|
|
vec2 v16 = unpackUnorm2x16(uy);
|
|
uint u17 = packHalf2x16(v2b);
|
|
vec2 v18 = unpackHalf2x16(uy);
|
|
|
|
// not present
|
|
noise2(v18); // ERROR, not present
|
|
|
|
float t__; // ERROR, no __ until revision 310
|
|
|
|
// ERROR, no __ until revision 310
|
|
#define __D
|
|
}
|