You can not select more than 25 topics
Topics must start with a letter or number, can include dashes ('-') and can be up to 35 characters long.
77 lines
1.9 KiB
77 lines
1.9 KiB
#version 450 core
|
|
#extension GL_KHR_memory_scope_semantics : enable
|
|
#extension GL_NV_cooperative_matrix : enable
|
|
#extension GL_EXT_shader_explicit_arithmetic_types_float16 : enable
|
|
|
|
layout (local_size_x = 64, local_size_y = 1, local_size_z = 1) in;
|
|
|
|
float<16> ftemplate16;
|
|
|
|
fcoopmatNV fnoparams;
|
|
|
|
fcoopmatNV<8, gl_ScopeSubgroup, 8, 8> fbadbits;
|
|
|
|
fcoopmatNV<16, gl_ScopeSubgroup, 8> fbadnumparams;
|
|
|
|
int X = 8;
|
|
|
|
fcoopmatNV<16, gl_ScopeSubgroup, 8, X> fbadparam;
|
|
|
|
layout(constant_id = 0) int Y = 1;
|
|
|
|
shared fcoopmatNV<16, gl_ScopeSubgroup, 16, 16> sharedmat;
|
|
|
|
layout(set = 0, binding = 0) buffer InvBlock {
|
|
fcoopmatNV<16, gl_ScopeSubgroup, 16, 16> bufmat;
|
|
} invblock;
|
|
|
|
void main()
|
|
{
|
|
fcoopmatNV<32, gl_ScopeSubgroup, 16, 8> f32_16_8;
|
|
fcoopmatNV<16, gl_ScopeSubgroup, 16, 8> f16_16_8;
|
|
|
|
// invalid implicit conversions
|
|
f32_16_8 = f16_16_8;
|
|
f32_16_8 = f16_16_8 + f16_16_8;
|
|
|
|
fcoopmatNV<16, gl_ScopeSubgroup, 8, 8> f16_8_8;
|
|
|
|
// mismatching dimensions
|
|
f16_16_8 = f16_8_8;
|
|
|
|
fcoopmatNV<16, gl_ScopeSubgroup, 8, Y> f16_8_Y;
|
|
fcoopmatNV<16, gl_ScopeSubgroup, 8, (Y+1)> f16_8_Y1;
|
|
|
|
// mismatching dimensions with specialization constants
|
|
f16_8_Y = f16_8_Y1;
|
|
|
|
// wrong arguments for constructor
|
|
f16_8_8 = fcoopmatNV<16, gl_ScopeSubgroup, 8, 8>(1, 1);
|
|
|
|
// can't construct from a builtin type
|
|
mat4 m4;
|
|
fcoopmatNV<32, gl_ScopeSubgroup, 4, 4> f32_4_4 = fcoopmatNV<32, gl_ScopeSubgroup, 4, 4>(m4);
|
|
|
|
// only support a single array subscript
|
|
f16_16_8[0][0];
|
|
|
|
// don't support scalar component selection
|
|
f16_16_8.x;
|
|
|
|
f16_16_8 * f16_16_8;
|
|
|
|
f16_16_8 + 1.0;
|
|
f16_16_8 - 1.0;
|
|
f16_16_8 / 1.0;
|
|
f16_16_8 += 1.0;
|
|
f16_16_8 -= 1.0;
|
|
f16_16_8 /= 1.0;
|
|
|
|
f16_16_8*2.0;
|
|
2.0*f16_16_8;
|
|
f32_16_8*float16_t(2.0);
|
|
float16_t(2.0)*f32_16_8;
|
|
|
|
transpose(f16_8_8);
|
|
}
|