[PATCH 4/5] d3dx9_36: Implemented ID3DXConstantTable_SetMatrix and ID3DXConstantTable_SetMatrixArray

[Michael Stefaniuc]

On 01/22/2011 12:22 AM, Travis Athougies wrote:
> matrix + i will advance by the size of 1 D3DXMATRIX, since matrix is a
> pointer to a D3DXMATRIX. Michael is right, I thought the code was
> clearer with the pointer arithmetic, but I now see more people are
> familiar with the array indexing style.
Actually the pointer cast draw my attention. Pointer casts are bad. Bad 
and evil. They serve to make the compiler stfu and for the code reader 
they are a big WARNING comment "There might be dragons!". And there was 
a dragon: the cast of a struct pointer to a pointer to its first field.
The pointer arithmetic to array conversion I have seen only after you 
explained it that way to Matteo.

bye
	michael

> On Fri, Jan 21, 2011 at 8:47 AM, Matteo Bruni<matteo.mystral at gmail.com>  wrote:
>> 2011/1/21 Michael Stefaniuc<mstefani at redhat.com>:
>>> On 01/21/2011 06:56 AM, Travis Athougies wrote:
>>>> On Thu, Jan 20, 2011 at 7:27 AM, Matteo Bruni<matteo.mystral at gmail.com>
>>>>   wrote:
>>>>>
>>>>> 2011/1/20 Travis Athougies<iammisc at gmail.com>:
>>>>>>
>>>>>> +            /* D3DXMATRIX is a union, one of whose elements is an
>>>>>> array, so it can be cast to a float pointer */
>>>>>> +           if (is_vertex_shader(This->desc.Version))
>>>>>> +               IDirect3DDevice9_SetVertexShaderConstantF(device,
>>>>>> desc.RegisterIndex + i, (float *)(matrix + i),
>>>>>> +                        desc.RegisterCount);
>>>>>> +           else
>>>>>> +               IDirect3DDevice9_SetPixelShaderConstantF(device,
>>>>>> desc.RegisterIndex + i, (float *)(matrix + i),
>>>>>> +                        desc.RegisterCount);
>>>>>
>>>>> Can't you just pass matrix->m[i] as parameter, instead of that cast?
>>>>>
>>>>
>>>> Uh no. If I were to do that, matrix->m[1] would be the second row, not
>>>> the second matrix. I'm trying to get at the second matrix. To
>>>> illustrate this, suppose the matrix were at 0x8000 (not going to
>>>> happen, but just pretend). matrix->m[1] would be at 0x8010, since
>>>> floats are 4 bytes. However, the next matrix is actually at 0x8040
>>>> (since sizeof(float) * 16 [the number of elements in the matrix] =
>>>> 64).
>>>
>>> So you want a pointer to the first element of the "i"th matrix, right?
>>>     &matrix[i]._11
>>> No cast, no explicit pointer arithmetic.
>>>
>>
>> Oh, right, I misread that (and Michael's suggestion is good). But then
>> I would expect the register index to be incremented by
>> desc.RegisterCount at each iteration. RegisterCount is supposedly 4,
>> but what if it is not? This calls for some SetMatrixArray tests.