81 lines
2.4 KiB
Plaintext
81 lines
2.4 KiB
Plaintext
#pragma clang diagnostic ignored "-Wmissing-prototypes"
|
|
|
|
#include <metal_stdlib>
|
|
#include <simd/simd.h>
|
|
|
|
using namespace metal;
|
|
|
|
typedef packed_float3 packed_rm_float3x2[2];
|
|
|
|
struct SSBOCol
|
|
{
|
|
float3x2 col_major0;
|
|
float3x2 col_major1;
|
|
};
|
|
|
|
struct SSBORow
|
|
{
|
|
packed_rm_float3x2 row_major0;
|
|
packed_rm_float3x2 row_major1;
|
|
};
|
|
|
|
void load_store_to_variable_col_major(device SSBOCol& v_29)
|
|
{
|
|
float3x2 loaded = v_29.col_major0;
|
|
v_29.col_major1 = loaded;
|
|
}
|
|
|
|
void load_store_to_variable_row_major(device SSBORow& v_41)
|
|
{
|
|
float3x2 loaded = transpose(float3x2(float3(v_41.row_major0[0]), float3(v_41.row_major0[1])));
|
|
v_41.row_major0[0] = float3(loaded[0][0], loaded[1][0], loaded[2][0]);
|
|
v_41.row_major0[1] = float3(loaded[0][1], loaded[1][1], loaded[2][1]);
|
|
}
|
|
|
|
void copy_col_major_to_col_major(device SSBOCol& v_29)
|
|
{
|
|
v_29.col_major0 = v_29.col_major1;
|
|
}
|
|
|
|
void copy_col_major_to_row_major(device SSBOCol& v_29, device SSBORow& v_41)
|
|
{
|
|
v_41.row_major0[0] = float3(v_29.col_major0[0][0], v_29.col_major0[1][0], v_29.col_major0[2][0]);
|
|
v_41.row_major0[1] = float3(v_29.col_major0[0][1], v_29.col_major0[1][1], v_29.col_major0[2][1]);
|
|
}
|
|
|
|
void copy_row_major_to_col_major(device SSBOCol& v_29, device SSBORow& v_41)
|
|
{
|
|
v_29.col_major0 = transpose(float3x2(float3(v_41.row_major0[0]), float3(v_41.row_major0[1])));
|
|
}
|
|
|
|
void copy_row_major_to_row_major(device SSBORow& v_41)
|
|
{
|
|
v_41.row_major0[0] = float3x2(float2(v_41.row_major1[0]), float2(v_41.row_major1[1]), float2(v_41.row_major1[2]))[0];
|
|
v_41.row_major0[1] = float3x2(float2(v_41.row_major1[0]), float2(v_41.row_major1[1]), float2(v_41.row_major1[2]))[1];
|
|
}
|
|
|
|
void copy_columns(device SSBOCol& v_29, device SSBORow& v_41)
|
|
{
|
|
v_29.col_major0[1] = float2(v_41.row_major0[0][1], v_41.row_major0[1][1]);
|
|
device float2*(v_41.row_major0[0][1], v_41.row_major0[1][1]) = v_29.col_major0[1];
|
|
}
|
|
|
|
void copy_elements(device SSBOCol& v_29, device SSBORow& v_41)
|
|
{
|
|
v_29.col_major0[0].y = v_41.row_major0[1u][0];
|
|
v_41.row_major0[1u][0] = v_29.col_major0[0].y;
|
|
}
|
|
|
|
kernel void main0(device SSBOCol& v_29 [[buffer(0)]], device SSBORow& v_41 [[buffer(1)]])
|
|
{
|
|
load_store_to_variable_col_major(v_29);
|
|
load_store_to_variable_row_major(v_41);
|
|
copy_col_major_to_col_major(v_29);
|
|
copy_col_major_to_row_major(v_29, v_41);
|
|
copy_row_major_to_col_major(v_29, v_41);
|
|
copy_row_major_to_row_major(v_41);
|
|
copy_columns(v_29, v_41);
|
|
copy_elements(v_29, v_41);
|
|
}
|
|
|