make_float4(x,y,z,w) but reads as (w,x,y,z)

Accelerated Computing CUDA CUDA Programming and Performance
1

EDIT: I realized the issue is that the make_float4() and make_int4() function lists the values as {x,y,z,w}, but when you read from a float4 or int4 type is lists the values as {w,x,y,z}.

What is going on here? This caused a bug because it was assumed that the ordering would stay the same as the constructor function.

2

Not sure what you mean. Here is the implementation of this function in header file vector_functions.h:

static __inline__ __host__ __device__ float4 make_float4(float x, float y, float z, float w)
{
  float4 t; t.x = x; t.y = y; t.z = z; t.w = w; return t;
}

The components are assigned to the matching components of the struct as expected. The struct layout can be found in header file vector_types.h:

struct __device_builtin__ __builtin_align__(16) float4
{
    float x, y, z, w;
};

Nothing surprising there either: x is at the lowest address, w at the highest.

3

The problem actually is from the VisualStudio intellisense, which lists the values as {w,x,y,z}, (in alphabetical order?).

Because I saw that I assumed that was the way they were laid out.

Here is a link to imgur of the screenshot:

[url]http://i.imgur.com/y4LloNX.png[/url]

4

I can see how this could be very confusing. Isn’t there an option to display the members of struct in “native” rather than alphabetical order?

5

There probably is a way, I just got temporarily puzzled by that. All is fixed now, and will look up how to change that VisualStudio setting.

Thanks.