OpSRem is converted to MSL as an unsigned operation

[giomasce]

Consider the attached shader, which has a OpSRem operation. When converted to MSL it becomes:

#include <metal_stdlib>
#include <simd/simd.h>

using namespace metal;

struct push_cb_struct
{
    float4 cb0[1];
};

struct main0_out
{
    float4 o0 [[color(0)]];
};

fragment main0_out main0(constant push_cb_struct& push_cb [[buffer(0)]])
{
    main0_out out = {};
    float _25 = precise::max(push_cb.cb0[0u].x, -2147483648.0);
    uint _32 = isnan(push_cb.cb0[0u].x) ? 0u : ((_25 >= 2147483648.0) ? 2147483647u : uint(int(_25)));
    float _34 = precise::max(push_cb.cb0[0u].y, -2147483648.0);
    uint _39 = isnan(push_cb.cb0[0u].y) ? 0u : ((_34 >= 2147483648.0) ? 2147483647u : uint(int(_34)));
    out.o0.x = float(int((_39 != 0u) ? (_32 - _39 * (_32 / _39)) : 4294967295u));
    out.o0.y = 0.0;
    out.o0.z = 0.0;
    out.o0.w = 0.0;
    float4 r0;
    return out;
}

The remainder operation is executed as unsigned instead of signed, which means that it is incorrect for negative numbers.