https://gcc.gnu.org/bugzilla/show_bug.cgi?id=99517
Bug ID: 99517
Summary: __builtin_convertvector with casts
Product: gcc
Version: 10.2.1
Status: UNCONFIRMED
Severity: normal
Priority: P3
Component: c
Assignee: unassigned at gcc dot gnu.org
Reporter: me+gcc.gnu at lelf dot lu
Target Milestone: ---
#include<stdint.h>
typedef int8_t GV __attribute__((vector_size(32),aligned(1)));
typedef int32_t IV __attribute__((vector_size(16),aligned(1)));
typedef int64_t JV __attribute__((vector_size(32),aligned(1)));
typedef double FV __attribute__((vector_size(32),aligned(1)));
GV FI(IV x){return(GV)__builtin_convertvector(x,FV);}
GV JI(IV x){return(GV)__builtin_convertvector(x,JV);}
x86_64
-Os or -O2
$ gcc-11x -Os -c cv.c -mavx2
FI is ok
0000000000000000 <FI>:
0: c5 fa e6 c8 vcvtdq2pd %xmm0,%xmm1
4: c5 f9 70 c0 ee vpshufd $0xee,%xmm0,%xmm0
9: c5 fa e6 c0 vcvtdq2pd %xmm0,%xmm0
d: c4 e3 75 18 c0 01 vinsertf128 $0x1,%xmm0,%ymm1,%ymm0
13: c3 retq
but JI is wrong (body of FI)
0000000000000014 <JI>:
14: c5 fa e6 c8 vcvtdq2pd %xmm0,%xmm1
18: c5 f9 70 c0 ee vpshufd $0xee,%xmm0,%xmm0
1d: c5 fa e6 c0 vcvtdq2pd %xmm0,%xmm0
21: c4 e3 75 18 c0 01 vinsertf128 $0x1,%xmm0,%ymm1,%ymm0
27: c3 retq
versions:
gcc-10x (GCC) 10.2.1 20210227
gcc-11x (GCC) 11.0.1 20210228 (experimental)
