Hi,
debugging a kernel I recognized that [font=“Courier New”]max_data[/font] and [font=“Courier New”]max_data2[/font] from the following
snippet are different pointers.
[codebox] extern shared float sdata
float * max_data = &sdata[N];
float * max_data2 = sdata + N * sizeof(float);[/codebox]
in this case [font=“Courier New”]max_data2[/font] is the same as [font=“Courier New”]&sdata[4*N][/font] which I do not understand.
I tried to use google to find some answers concerning “cuda pointer arithmetics”, but
found nothing usable.
Greetings, Uwe
Hi,
debugging a kernel I recognized that [font=“Courier New”]max_data[/font] and [font=“Courier New”]max_data2[/font] from the following
snippet are different pointers.
[codebox] extern shared float sdata
float * max_data = &sdata[N];
float * max_data2 = sdata + N * sizeof(float);[/codebox]
in this case [font=“Courier New”]max_data2[/font] is the same as [font=“Courier New”]&sdata[4*N][/font] which I do not understand.
You don’t need to do * sizeof(float) - this is actually equivalent to doing
float * max_data2 = &sdata[N * sizeof(float)];
if I’m not mistaken
The pointer type enforces proper arithmetic, incrementing only by meaningful amounts - if you really wanted to advance the pointer by one byte at a time (instead of one float at a time), you’d have to cast it to char*
You don’t need to do * sizeof(float) - this is actually equivalent to doing
float * max_data2 = &sdata[N * sizeof(float)];
if I’m not mistaken
The pointer type enforces proper arithmetic, incrementing only by meaningful amounts - if you really wanted to advance the pointer by one byte at a time (instead of one float at a time), you’d have to cast it to char*
Oh holy s**t, this is something I should have knowm myself. Debugging for some hours made my mind dump.
Thanks & Greetings,
Uwe