mirror of
https://github.com/KhronosGroup/glslang
synced 2024-11-15 06:10:05 +00:00
252 lines
6.2 KiB
GLSL
252 lines
6.2 KiB
GLSL
#version 450
|
|
|
|
#extension GL_EXT_shader_explicit_arithmetic_types: enable
|
|
#extension GL_EXT_shader_explicit_arithmetic_types_int8: require
|
|
#extension GL_EXT_shader_explicit_arithmetic_types_int16: require
|
|
#extension GL_EXT_shader_explicit_arithmetic_types_int32: require
|
|
#extension GL_EXT_shader_explicit_arithmetic_types_int64: require
|
|
#extension GL_EXT_shader_explicit_arithmetic_types_float16: require
|
|
#extension GL_EXT_shader_explicit_arithmetic_types_float32: require
|
|
#extension GL_EXT_shader_explicit_arithmetic_types_float64: require
|
|
|
|
layout(binding = 0) uniform Uniforms
|
|
{
|
|
uint index;
|
|
};
|
|
|
|
layout(std140, binding = 1) uniform Block
|
|
{
|
|
int16_t i16;
|
|
i16vec2 i16v2;
|
|
i16vec3 i16v3;
|
|
i16vec4 i16v4;
|
|
uint16_t u16;
|
|
u16vec2 u16v2;
|
|
u16vec3 u16v3;
|
|
u16vec4 u16v4;
|
|
} block;
|
|
|
|
void main()
|
|
{
|
|
}
|
|
|
|
void literal()
|
|
{
|
|
const int16_t i16Const[3] =
|
|
{
|
|
int16_t(-0x1111), // Hex
|
|
int16_t(-1), // Dec
|
|
int16_t(040000), // Oct
|
|
};
|
|
|
|
int16_t i16 = i16Const[index];
|
|
|
|
const uint16_t u16Const[] =
|
|
{
|
|
uint16_t(0xFFFF), // Hex
|
|
uint16_t(65535), // Dec
|
|
uint16_t(077777), // Oct
|
|
};
|
|
|
|
uint16_t u16 = u16Const[index];
|
|
}
|
|
|
|
void typeCast16()
|
|
{
|
|
i8vec2 i8v;
|
|
u8vec2 u8v;
|
|
i16vec2 i16v;
|
|
u16vec2 u16v;
|
|
i32vec2 i32v;
|
|
u32vec2 u32v;
|
|
i64vec2 i64v;
|
|
u64vec2 u64v;
|
|
f16vec2 f16v;
|
|
f32vec2 f32v;
|
|
f64vec2 f64v;
|
|
bvec2 bv;
|
|
|
|
i32v = i16v; // int16_t -> int32_t
|
|
i32v = u16v; // uint16_t -> int32_t
|
|
u16v = i16v; // int16_t -> uint16_t
|
|
u32v = i16v; // int16_t -> uint32_t
|
|
i64v = i16v; // int16_t -> int64_t
|
|
u64v = i16v; // int16_t -> uint64_t
|
|
u32v = u16v; // uint16_t -> uint32_t
|
|
i64v = u16v; // uint16_t -> int64_t
|
|
u64v = u16v; // uint16_t -> uint64_t
|
|
f16v = i16v; // int16_t -> float16_t
|
|
f32v = i16v; // int16_t -> float32_t
|
|
f64v = i16v; // int16_t -> float64_t
|
|
f16v = u16v; // uint16_t -> float16_t
|
|
f32v = u16v; // uint16_t -> float32_t
|
|
f64v = u16v; // uint16_t -> float64_t
|
|
|
|
i32v = i32vec2(i16v); // int16_t -> int32_t
|
|
i32v = i32vec2(u16v); // uint16_t -> int32_t
|
|
u16v = u16vec2(i16v); // int16_t -> uint16_t
|
|
u32v = u32vec2(i16v); // int16_t -> uint32_t
|
|
i64v = i64vec2(i16v); // int16_t -> int64_t
|
|
u64v = i64vec2(i16v); // int16_t -> uint64_t
|
|
u32v = u32vec2(u16v); // uint16_t -> uint32_t
|
|
i64v = i64vec2(u16v); // uint16_t -> int64_t
|
|
u64v = i64vec2(u16v); // uint16_t -> uint64_t
|
|
f16v = f16vec2(i16v); // int16_t -> float16_t
|
|
f32v = f32vec2(i16v); // int16_t -> float32_t
|
|
f64v = f64vec2(i16v); // int16_t -> float64_t
|
|
f16v = f16vec2(u16v); // uint16_t -> float16_t
|
|
f32v = f32vec2(u16v); // uint16_t -> float32_t
|
|
f64v = f64vec2(u16v); // uint16_t -> float64_t
|
|
|
|
i8v = i8vec2(i16v); // int16_t -> int8_t
|
|
i8v = i8vec2(u16v); // uint16_t -> int8_t
|
|
u8v = u8vec2(i16v); // int16_t -> uint8_t
|
|
u8v = u8vec2(u16v); // uint16_t -> uint8_t
|
|
i16v = u8vec2(u16v); // uint16_t -> int16_t
|
|
i16v = i16vec2(bv); // bool -> int16
|
|
u16v = u16vec2(bv); // bool -> uint16
|
|
bv = bvec2(i16v); // int16 -> bool
|
|
bv = bvec2(u16v); // uint16 -> bool
|
|
}
|
|
void operators()
|
|
{
|
|
u16vec3 u16v;
|
|
int16_t i16;
|
|
uvec3 uv;
|
|
int32_t i;
|
|
int64_t i64;
|
|
bool b;
|
|
|
|
// Unary
|
|
u16v++;
|
|
i16--;
|
|
++i16;
|
|
--u16v;
|
|
|
|
u16v = ~u16v;
|
|
|
|
i16 = +i16;
|
|
u16v = -u16v;
|
|
|
|
// Arithmetic
|
|
i16 += i16;
|
|
u16v -= u16v;
|
|
i *= i16;
|
|
uv /= u16v;
|
|
uv %= i16;
|
|
|
|
uv = u16v + uv;
|
|
i64 = i16 - i64;
|
|
uv = u16v * uv;
|
|
i64 = i16 * i64;
|
|
i = i16 % i;
|
|
|
|
// Shift
|
|
u16v <<= i16;
|
|
i16 >>= u16v.y;
|
|
|
|
i16 = i16 << u16v.z;
|
|
uv = u16v << i;
|
|
|
|
// Relational
|
|
b = (u16v.x != i16);
|
|
b = (i16 == u16v.x);
|
|
b = (u16v.x > uv.y);
|
|
b = (i16 < i);
|
|
b = (u16v.y >= uv.x);
|
|
b = (i16 <= i);
|
|
|
|
// Bitwise
|
|
uv |= i16;
|
|
i = i16 | i;
|
|
i64 &= i16;
|
|
uv = u16v & uv;
|
|
uv ^= i16;
|
|
u16v = u16v ^ i16;
|
|
}
|
|
|
|
void builtinFuncs()
|
|
{
|
|
i16vec2 i16v;
|
|
i16vec4 i16v4;
|
|
u16vec3 u16v;
|
|
u16vec2 u16v2;
|
|
u16vec4 u16v4;
|
|
bvec3 bv;
|
|
int16_t i16;
|
|
uint16_t u16;
|
|
int32_t i32;
|
|
uint32_t u32;
|
|
int64_t i64;
|
|
uint64_t u64;
|
|
|
|
// abs()
|
|
i16v = abs(i16v);
|
|
|
|
// sign()
|
|
i16 = sign(i16);
|
|
|
|
// min()
|
|
i16v = min(i16v, i16);
|
|
i16v = min(i16v, i16vec2(-1));
|
|
u16v = min(u16v, u16);
|
|
u16v = min(u16v, u16vec3(0));
|
|
|
|
// max()
|
|
i16v = max(i16v, i16);
|
|
i16v = max(i16v, i16vec2(-1));
|
|
u16v = max(u16v, u16);
|
|
u16v = max(u16v, u16vec3(0));
|
|
|
|
// clamp()
|
|
i16v = clamp(i16v, -i16, i16);
|
|
i16v = clamp(i16v, -i16v, i16v);
|
|
u16v = clamp(u16v, -u16, u16);
|
|
u16v = clamp(u16v, -u16v, u16v);
|
|
|
|
// mix()
|
|
i16 = mix(i16v.x, i16v.y, true);
|
|
i16v = mix(i16vec2(i16), i16vec2(-i16), bvec2(false));
|
|
u16 = mix(u16v.x, u16v.y, true);
|
|
u16v = mix(u16vec3(u16), u16vec3(-u16), bvec3(false));
|
|
|
|
//pack
|
|
i32 = pack32(i16v);
|
|
i64 = pack64(i16v4);
|
|
u32 = pack32(u16v2);
|
|
u64 = pack64(u16v4);
|
|
|
|
i16v = unpack16(i32);
|
|
i16v4 = unpack16(i64);
|
|
u16v2 = unpack16(u32);
|
|
u16v4 = unpack16(u64);
|
|
|
|
// lessThan()
|
|
bv = lessThan(u16v, u16vec3(u16));
|
|
bv.xy = lessThan(i16v, i16vec2(i16));
|
|
|
|
// lessThanEqual()
|
|
bv = lessThanEqual(u16v, u16vec3(u16));
|
|
bv.xy = lessThanEqual(i16v, i16vec2(i16));
|
|
|
|
// greaterThan()
|
|
bv = greaterThan(u16v, u16vec3(u16));
|
|
bv.xy = greaterThan(i16v, i16vec2(i16));
|
|
|
|
// greaterThanEqual()
|
|
bv = greaterThanEqual(u16v, u16vec3(u16));
|
|
bv.xy = greaterThanEqual(i16v, i16vec2(i16));
|
|
|
|
// equal()
|
|
bv = equal(u16v, u16vec3(u16));
|
|
bv.xy = equal(i16v, i16vec2(i16));
|
|
|
|
// notEqual()
|
|
bv = notEqual(u16v, u16vec3(u16));
|
|
bv.xy = notEqual(i16v, i16vec2(i16));
|
|
}
|
|
|
|
// Type conversion for specialization constant
|
|
layout(constant_id = 100) const int16_t si16 = int16_t(-10);
|
|
layout(constant_id = 101) const uint16_t su16 = uint16_t(20);
|