I am currently running Redhat 9 with gcc version 3.2.2. I am able to see the use of xmm registers in the listing, but never in packed mode. I tried
float slope_lut_h, slope_lut_v, temp __attribute__ ((vector_size (16))); temp = slope_lut_h + slope_lut_v;
only to get an illegal operands error message. Also, how does one assign a value (the same value 4 times) to such a float vector?
I also tried
__builtin_prefetch (&delta_ptr[i+16], 1, 1);
with the result Illegal instruction (core dumped). This is a P4, and may not support the cache prefetch. Does gcc have a function to return the processor capability mask?
I have used
typedef int v4sf __attribute__ ((mode(V4SF))); v4sf matrix[256][256] __attribute__ ((aligned (16)));
as an example of my array declarations.
Does the 3.2.2 compiler support the sse packed float instructions? Is there a site with some example code I could examine.
Thanks in advance for any advice or pointers. I am hesitant to join the gcc list, because I know little about writing compilers. I do like to push the envelope in applications code, hopefully without using assembler codes.
Thanks again, don
-- ------------------------------------------------------------------- | Don Fisher dfisher@xxxxxxxxxxxxxx | | Steward Observatory | | 933 N. Cherry Ave. VOICE: (520)621-7647 | | University of Arizona FAX: (520)621-9843 | | Tucson, AZ 85721 | -------------------------------------------------------------------
