https://gcc.gnu.org/bugzilla/show_bug.cgi?id=124133
Avinash Jayakar <avinashd at gcc dot gnu.org> changed:
What |Removed |Added
----------------------------------------------------------------------------
CC| |avinashd at gcc dot gnu.org
--- Comment #10 from Avinash Jayakar <avinashd at gcc dot gnu.org> ---
There was an ICE with the following code as well:
#include <altivec.h>
typedef unsigned char uint8_t;
typedef signed long int64_t;
typedef struct {
float d;
uint8_t qs[32 / 2];
} block_q4_0;
template <typename TA>
class tinyBLAS_Q0_PPC {
public:
int64_t k;
tinyBLAS_Q0_PPC(int64_t k)
: k(k) {
}
private:
void packNormal_q4_fp16(const block_q4_0 * a) {
vector signed char v_qs = (vector signed char)vec_xl(0, a->qs);
}
};
found when building llama.cpp with gcc15. Bisect results gave the same bad
commit i.e., r14-4793-gdad311874ac3b3cf4eca1c04f67cae80c953f7b8
Reproduction: https://godbolt.org/z/Gf4v47qKE
Please let me know if a separate bug is needed for it.
