b3f6e3de8e
The algorithm was too conservative causing lots of unnecessary temporaries to be created.
40 lines
667 B
Plaintext
40 lines
667 B
Plaintext
#include <metal_stdlib>
|
|
#include <simd/simd.h>
|
|
|
|
using namespace metal;
|
|
|
|
struct SSBO
|
|
{
|
|
float4x4 mvp;
|
|
float4 in_data[1];
|
|
};
|
|
|
|
struct SSBO2
|
|
{
|
|
float4 out_data[1];
|
|
};
|
|
|
|
kernel void main0(uint3 gl_GlobalInvocationID [[thread_position_in_grid]], device SSBO& _28 [[buffer(0)]], device SSBO2& _52 [[buffer(1)]])
|
|
{
|
|
int i = 0;
|
|
float4 _56;
|
|
_56 = _28.in_data[gl_GlobalInvocationID.x];
|
|
float4 _42;
|
|
for (;;)
|
|
{
|
|
_42 = _28.mvp * _56;
|
|
i++;
|
|
if (i < 16)
|
|
{
|
|
_56 = _42;
|
|
continue;
|
|
}
|
|
else
|
|
{
|
|
break;
|
|
}
|
|
}
|
|
_52.out_data[gl_GlobalInvocationID.x] = _42;
|
|
}
|
|
|